A liquor quality evaluation method and system based on multi-source data analysis

Abstract

The present application relates to the technical field of liquor brewing and quality detection, and particularly relates to a liquor quality evaluation method and system based on multi-source data analysis. The liquor quality evaluation method based on multi-source data analysis comprises the following steps: S1: obtaining multi-source data information of liquor according to multiple detection and analysis methods, and performing preprocessing operation on the multi-source data information, wherein the multi-source data information comprises raw material data, process data, product data and external data; S2: performing multi-dimensional feature extraction based on the processed product data, and constructing a fusion database through the extracted multi-dimensional feature data and the processed multi-source data information. Through integrated and automated design, the present application integrates the detection equipment originally scattered in multiple laboratories into a set of high-efficiency operating system, realizes automatic acquisition of full-process data flow from raw material detection to finished product analysis, greatly improves the detection efficiency and effectively reduces the human operation error.

Description

Technical Field

[0001] This invention relates to the field of liquor brewing and quality testing technology, and in particular to a liquor quality evaluation method and system based on multi-source data analysis. Background Technology

[0002] Currently, the evaluation of baijiu quality still relies heavily on manual sensory assessment, resulting in a significant lack of objective quantitative standards. This is particularly true for baijiu brewed using non-traditional raw materials such as barley, fruit, and potatoes, whose flavor profiles differ significantly from traditional sorghum-based baijiu. This renders existing quality evaluation systems based on sorghum-based characteristics completely ineffective. Current industry testing methods are limited to end-product analysis, lacking the ability to collect comprehensive data covering raw material characteristics, process parameters, and dynamic changes in flavor compounds, and are unable to achieve multi-dimensional data correlation modeling and intelligent analysis.

[0003] Therefore, there is an urgent need to develop a method and system for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis. This system should possess full-process data fusion capabilities, integrating multi-source sensors for raw material component detection, process parameter monitoring, and dynamic tracking of flavor substances to construct a digital evaluation model covering the entire chain from raw materials to finished products. The system needs to overcome the limitations of traditional single-dimensional evaluation, utilizing machine learning algorithms to achieve correlation analysis between flavor substances and process parameters, forming an objective quantitative evaluation system applicable to baijiu made from non-traditional raw materials, and providing the industry with standardized and intelligent quality control solutions. Summary of the Invention

[0004] To overcome the shortcomings of existing baijiu quality evaluation methods, which rely heavily on human sensory evaluation and lack objective quantitative standards, especially given the significant differences in flavor profiles of baijiu made from non-traditional raw materials leading to the failure of traditional evaluation systems, and the limitations of existing testing methods that are limited to end-product analysis and cannot achieve multi-dimensional data correlation modeling and intelligent analysis throughout the entire process, this invention provides a baijiu quality evaluation method and system based on multi-source data analysis.

[0005] Technical solution: A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis, comprising the following steps: S1: Obtain multi-source data information of liquor based on various detection and analysis methods, and perform preprocessing operations on the multi-source data information, which includes raw material data, process data, product data and external data; S2: Multidimensional feature extraction is performed based on the processed product data, and a fusion database is constructed using the extracted multidimensional feature data and processed multi-source data information. The fusion database includes wine sample information, raw material feature information, process feature information, flavor substance information, metabolite information, and sensory flavor score. S3: Based on the raw material characteristic information, process characteristic information and flavor substance information in the constructed fusion database, a causal graph is generated by executing a causal discovery algorithm. The causal graph is a directed acyclic graph. S4: Based on the critical paths identified in the causal graph, a regression algorithm is used to obtain the path coefficients of the critical paths. The total effect from the raw material node to the flavor substance node through all direct and indirect paths is calculated using the path coefficients. The contribution of the raw material node is quantified based on the total effect. S5: Based on the metabolite information in the fusion database, select compounds that have been proven to have health benefits by literature or cell experiments, and calculate the health value index through beneficial metabolic data; S6: Generate a visual report based on the sensory flavor score, causal graph, contribution, and health value index.

[0006] Preferably, the process of acquiring multi-source data information on baijiu using various detection and analysis methods, and preprocessing the multi-source data information, includes raw material data, process data, product data, and external data. Specifically, the multi-source data information acquired using various detection and analysis methods includes: raw material data acquired through near-infrared spectroscopy, high-performance liquid chromatography, ultraviolet spectrophotometry, and biochemical analyzers; process data acquired through temperature sensors, humidity sensors, pH sensors, and data acquisition devices; flavor substance data acquired through gas chromatography-mass spectrometry; metabolic data acquired through liquid chromatography-high-resolution mass spectrometry; sensory evaluation data acquired through a tasting panel and standardized scoring sheets; and data acquired through market research. External data obtained through research, web scraping, and sales data analysis; the preprocessing operations include handling missing values, removing outliers, time series alignment, and data standardization; handling missing values ​​refers to missing data caused by accidental sensor malfunction or detection failure, which is filled by interpolation or based on other samples from the same batch; removing outliers refers to using statistical methods to identify and remove abnormal detection values ​​that significantly deviate from the normal range; time series alignment refers to accurately mapping the time series data of process parameters to specific brewing batches; data standardization refers to processing continuous numerical data using Z-Score standardization and digitizing descriptive terms for sensory evaluation; the product data includes flavor substance data, metabolic data, and sensory evaluation data.

[0007] Preferably, the process involves multidimensional feature extraction based on processed product data, and the construction of a fusion database using the extracted multidimensional feature data and processed multi-source data. This fusion database includes wine sample information, raw material characteristic information, process characteristic information, flavor substance information, metabolite information, and sensory flavor scores. Specifically, the multidimensional feature extraction based on processed product data includes extracting the absolute concentration or relative peak area of ​​specific flavor compounds as features, summarizing flavor substance information based on these features, and comparing the metabolic data with the database after peak identification, peak alignment, and noise reduction using software. This process is used to identify and classify certain compounds. The process involves generating a metabolic feature matrix containing the peak intensities of all samples and detected compounds, and summarizing metabolite information based on this matrix. By averaging scores from multiple tasters and performing principal component analysis for dimensionality reduction, principal component scores representing core sensory dimensions are extracted as features, and sensory flavor scores are summarized based on these features. The construction of a fusion database using the extracted multidimensional feature data and processed multi-source data involves associating the flavor substance information, metabolite information, sensory flavor scores, and raw material and process feature information from the multi-source data through wine sample information, ultimately constructing the fusion database.

[0008] Preferably, the causal graph is generated by executing a causal discovery algorithm based on the raw material feature information, process feature information, and flavor substance information in the constructed fusion database. The causal graph is a directed acyclic graph, including: the causal discovery algorithm is an algorithm that infers the causal direction between variables based on assumptions and statistical principles. The main causal discovery algorithm used is the PC algorithm. The PC algorithm is based on conditional independence testing. Starting from a completely undirected graph, it gradually deletes edges through conditional independence testing to determine the undirected association between raw material features, process features, and flavor substance information. Then, it assigns directionality to the remaining connecting edges according to the causal direction determination rules, and finally obtains a directed acyclic graph to represent the causal relationship between raw material features, process features, and flavor substance information, and to show the causal influence path to form a causal graph; the nodes in the directed acyclic graph are each feature or flavor substance, and the edges represent causal relationships.

[0009] Preferably, the step of obtaining the path coefficients of the critical paths identified in the causal graph using a regression algorithm includes: the associated path is a node path from the raw material node to the flavor substance node in the causal graph, which includes the raw material node, the influencing node, and the flavor substance node; the influencing node is any node on the node path other than the raw material node and the flavor substance node; the regression algorithm is a formula used to obtain the path coefficients; the path coefficients directly measure the strength and direction of the influence between variables, and the formula for obtaining the path coefficients is: ; ; in To quantify and standardize data that affects nodes; This is the baseline value for predicting the dependent variable when the independent variable is 0; it is also called the intercept term. for right Path coefficients; Quantitative and standardized data for raw material nodes; This represents the error term in the regression equation; Quantitative and standardized data for flavor compound nodes; for right Path coefficients; for right Path coefficients.

[0010] Preferably, the step of calculating the total effect from the raw material node to the flavor substance node through all direct and indirect paths using the path coefficients includes: calculating the total effect between the raw material node and the flavor substance node using the path coefficients; the total effect includes a first direct effect between the raw material node and the flavor substance node and a first indirect effect between the influencing node and the flavor substance node, the first indirect effect including a second direct effect and a third direct effect, the second direct effect being the effect between the influencing node and the flavor substance node; the third direct effect being the effect between the raw material node and the influencing node, wherein the formula for calculating the total effect is: ; ; ; in This is the first direct effect between the raw material node and the flavor substance node; This is the first indirect effect affecting the connection between nodes and flavor compound nodes; The total effect of raw material nodes reaching flavor substance nodes for right Path coefficients; for right Path coefficients; for right Path coefficients.

[0011] Preferably, quantifying the contribution of the raw material node based on the total effect includes: the contribution represents the degree of influence of the raw material characteristics on the flavor compound, obtained by comparing the total effect of the raw material node on the flavor compound with the total effect of all significant influences on the flavor compound, wherein the formula for obtaining the contribution is: ; in for right The degree of contribution; for right The overall effect; For all significant effects The sum of the total effects of the variables; The value is expressed as a percentage.

[0012] Preferably, the step of screening compounds with health benefits confirmed by literature or cell experiments based on metabolite information in the fusion database, and calculating a health value index using beneficial metabolic data, includes: after obtaining metabolite information from the fusion database, screening for beneficial metabolites with health benefits confirmed by literature or cell experiments; directly obtaining the standardized content of the beneficial metabolites from the fusion database; then determining the half-maximal inhibitory concentration (IC50) of each beneficial metabolite using a DPPH free radical scavenging experiment; obtaining weights based on the ratio of the minimum IC50 value to the measured values ​​of different metabolites; and calculating the health value index using the standardized content of different beneficial metabolites and the weight allocation, wherein the formula for obtaining the health value index is: ; in As a health value index; For the first Standardized content of beneficial metabolites in wine samples; For the first The weight of beneficial metabolites; This is the summation symbol.

[0013] Preferably, the step of generating a visualization report based on the sensory flavor score, causal graph, contribution rate, and health value index includes: the visualization report directly displaying the product's performance across multiple sensory dimensions in the form of a radar chart using the sensory flavor score; displaying the main pathways affecting the product's key flavors using the causal graph; displaying the product's main components and their content using the contribution rate, and listing the top-ranking components by contribution rate and their contribution values; and displaying the HVI score and level using the health value index in the form of a dashboard.

[0014] A liquor quality evaluation system based on multi-source data analysis includes: The data acquisition module is responsible for systematically collecting four major categories of data to provide basic data for subsequent data processing and analysis. The data processing module cleans, aligns, and standardizes the collected multi-source data to improve data quality and eliminate the impact of different units and orders of magnitude on the model. The feature engineering module extracts multidimensional features from the raw data and constructs a feature matrix for subsequent causal discovery and machine learning modeling. The data fusion module creates a core database that links data from all dimensions together to form a structured wide table for subsequent causal discovery and machine learning modeling. The reasoning and calculation module uses a causal discovery algorithm to generate a causal graph, quantifies the contribution of each feature, and calculates the health value index, providing a basis for the quality analysis and optimization of baijiu. The evaluation output module integrates scores from multiple dimensions into a comprehensive index and generates a visual report, providing an intuitive reference for the production and decision-making of baijiu.

[0015] Compared with existing technologies, this invention has the following advantages: Through integrated and automated design, this invention integrates testing equipment originally scattered across multiple laboratories into a highly efficient system, achieving automated data acquisition throughout the entire process from raw material testing to finished product analysis. This significantly improves testing efficiency and effectively reduces human error. Simultaneously, with the help of a unified central data processing terminal, it can deeply correlate raw material characteristics, process parameters, and finished product flavor substance data, providing strong data support and hardware foundation for causal discovery algorithms and path total effect quantification. Furthermore, the innovative health value index calculation technology miniaturizes and automates cell experiments, seamlessly integrating them into the system, enabling rapid, in-situ assessment of the health attributes of baijiu. This groundbreaking design far surpasses the efficiency and accuracy of traditional laboratory models. Overall, this system is specifically customized for the component analysis and quality evaluation of non-traditional raw material baijiu, not only filling the gap in dedicated evaluation tools in this field but also providing strong technical support for the intelligent transformation of the baijiu industry with its high degree of specialization and practicality. Attached Figure Description

[0016] Figure 1 This is a flowchart of the liquor quality evaluation method based on multi-source data analysis of the present invention. Figure 2 This is a schematic diagram of the structure of the liquor quality evaluation system based on multi-source data analysis of the present invention. Detailed Implementation

[0017] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0018] Example 1: A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis, such as... Figure 1As shown, it includes the following steps: S1: Obtain multi-source data information of liquor using various detection and analysis methods, and perform preprocessing operations on the multi-source data information, which includes raw material data, process data, product data, and external data; The multi-source data information on baijiu obtained through various detection and analysis methods includes raw material data acquired by near-infrared spectroscopy, high-performance liquid chromatography, ultraviolet spectrophotometry, and biochemical analyzer; process data acquired by temperature sensors, humidity sensors, pH sensors, and data acquisition devices; flavor substance data acquired by gas chromatography-mass spectrometry; metabolic data acquired by liquid chromatography-high-resolution mass spectrometry; sensory evaluation data acquired by a panel of sommeliers and standardized scoring sheets; and external data acquired through market research, web crawling, and sales data analysis. The preprocessing operations include handling missing values ​​and removing outliers. The process includes constant values, time series alignment, and data standardization. Missing values ​​are filled by interpolation or based on other samples from the same batch, addressing data lost due to occasional sensor malfunction or detection failure. Outlier removal uses statistical methods to identify and remove abnormal detection values ​​that significantly deviate from the normal range. Time series alignment precisely maps the time series data of process parameters to specific brewing batches. Data standardization involves Z-score standardization of continuous numerical data and digital encoding of descriptive terms for sensory evaluation. Product data includes flavor compound data, metabolic data, and sensory evaluation data.

[0019] It should also be noted that in the multi-source data collection of liquor, an analysis system is constructed and complementary technical means are used to ensure the integrity of data dimensions and measurement accuracy.

[0020] Raw material characteristic analysis relies on a cluster of equipment including near-infrared spectroscopy, high-performance liquid chromatography, ultraviolet-visible spectrophotometer, and fully automated biochemical analyzer to achieve comprehensive characterization of raw material components. Brewing process parameters are dynamically acquired through a distributed sensor network integrating multiple sensors for temperature, humidity, and pH, along with a dedicated data acquisition terminal, to fully record the evolution of environmental parameters during fermentation.

[0021] Flavor component analysis employs gas chromatography-mass spectrometry (GC-MS) to achieve precise qualitative and quantitative analysis of complex flavor compounds through optimized separation and detection conditions. Metabolite analysis utilizes a liquid chromatography-high-resolution mass spectrometry (LC-HDMS) platform to screen and identify trace metabolites in the wine. Sensory characteristic data are obtained by a professionally trained tasting panel according to standardized evaluation procedures, forming a subjective quantitative description of the product's flavor characteristics. Simultaneously, external data on product market performance and consumer feedback are collected through structured market research, online data collection, and sales record analysis.

[0022] In the data preprocessing stage, a rigorous quality control process is established. For missing values ​​caused by equipment failure or operational errors, time-series-based interpolation algorithms or batch-based sample data imputation strategies are employed. Statistical process control methods are used to identify and remove outlier values ​​that significantly deviate from the normal fluctuation range, ensuring the reliability of the dataset. Through timestamp alignment technology, a precise correspondence is established between the time-series data of process parameters and specific production batches, constructing a complete production history archive.

[0023] In the data standardization phase, Z-score normalization was applied to continuous numerical variables to eliminate the impact of dimensional differences on the model; a digital coding system was established for descriptive terms in sensory evaluations to achieve the quantitative transformation of qualitative information. These processes significantly improved the accuracy and interpretability of subsequent data analysis.

[0024] The final product feature dataset contains complete quantitative information in three dimensions: flavor profile, metabolite composition, and sensory evaluation, laying a data foundation for building a comprehensive product quality evaluation system.

[0025] S2: Multidimensional feature extraction is performed based on the processed product data, and a fusion database is constructed by the extracted multidimensional feature data and the processed multi-source data information. The fusion database includes wine sample information, raw material feature information, process feature information, flavor substance information, metabolite information, and sensory flavor score. The multidimensional feature extraction based on processed product data includes extracting the absolute concentration or relative peak area of ​​specific flavor compounds as features, summarizing flavor substance information based on these features; performing peak identification, peak alignment, and noise reduction on metabolic data using software, comparing it with a database, qualitatively identifying some compounds, and generating a metabolic feature matrix containing the peak intensities of all samples and all detected compounds, summarizing metabolite information based on this metabolic feature matrix; averaging scores from multiple sommeliers and performing principal component analysis for dimensionality reduction to extract principal component scores representing core sensory dimensions as features, summarizing sensory flavor scores based on these features; and constructing a fusion database using the extracted multidimensional feature data and processed multi-source data information involves associating the flavor substance information, metabolite information, sensory flavor scores, and raw material and process feature information from the multi-source data information through wine sample information to ultimately construct the fusion database.

[0026] It should also be noted that, based on the completion of data preprocessing, the construction of multi-dimensional feature engineering was initiated. Differentiated feature extraction schemes were designed for different types of data characteristics to fully mine the product feature information contained in the data.

[0027] In terms of flavor component characterization, the absolute concentration of key flavor compounds and their relative peak areas in the chromatogram are selected as quantitative indicators. These screened characteristic parameters can objectively reflect the actual content levels and relative compositional relationships of various flavor components in the wine, providing reliable data support for establishing a scientific flavor substance archive.

[0028] For metabolomics data, a professional analysis platform was used to perform peak detection, peak alignment, and background noise filtering preprocessing on the raw spectra. Then, a metabolite database was used for compound structure comparison and identification, ultimately generating a standardized feature matrix containing the peak intensities of all sample compounds. This processing workflow effectively ensured the accuracy of qualitative and quantitative metabolite identification, laying the foundation for subsequent metabolic pathway analysis.

[0029] In terms of digitizing sensory characteristics, expert group scoring was averaged to eliminate individual evaluation bias. Principal component analysis (PCA) was introduced to reduce the dimensionality of multidimensional sensory indicators, including aroma, taste, and flavor and body characteristics. By extracting the principal component scores with the highest contribution rates as representative features, the main variation information of the original sensory data was preserved, while the sensory characteristics were simplified, significantly improving the usability of the sensory data. Finally, using the sample number as a link, flavor characteristics, metabolite characteristics, sensory scoring characteristics, raw material characteristics, and process parameters were integrated to construct a complete multi-source characteristic database for baijiu. This database not only achieves standardized storage and unified management of different types of data but also establishes a complete data foundation for subsequent quality traceability and in-depth flavor prediction analysis through the inherent correlation between data. This multidimensional feature fusion strategy preserves the independent characteristics of each data dimension while creating cross-data mining, opening up a new technical path for comprehensively understanding the quality characteristics of baijiu.

[0030] S3: Based on the raw material characteristic information, process characteristic information and flavor substance information in the constructed fusion database, a causal graph is generated by executing a causal discovery algorithm. The causal graph is a directed acyclic graph. The causal discovery algorithm described is an algorithm that infers the causal direction between variables based on assumptions and statistical principles. The main causal discovery algorithm used is the PC algorithm. The PC algorithm is based on conditional independence testing. Starting from a completely undirected graph, it gradually removes edges through conditional independence testing to determine the undirected association between raw material characteristics, process characteristics, and flavor substance information. Then, it assigns directionality to the remaining connecting edges according to the causal direction determination rules, and finally obtains a directed acyclic graph to represent the causal relationship between raw material characteristics, process characteristics, and flavor substance information, and to show the causal influence path to form a causal graph. The nodes in the directed acyclic graph are each characteristic or flavor substance, and the edges represent causal relationships.

[0031] It should also be noted that this invention employs a PC algorithm based on conditional independence testing for causal discovery. This method determines the causal direction between variables through statistical inference. Specifically, a fully connected undirected graph containing all raw material characteristics, process parameters, and flavor substance nodes is first constructed. Subsequently, based on the conditional independence test results, the algorithm gradually removes connections lacking statistical dependencies, initially establishing the network of relationships between variables. After determining the undirected graph structure, the remaining connections are assigned directionality according to causal direction determination rules, including time-series priority rules, conditional independence test rules, domain knowledge constraint rules, and model scoring rules, ultimately generating a directed acyclic graph with a clear causal orientation. The nodes in this graph structure represent different characteristic variables, and the directed edges represent the causal influence paths between variables, clearly demonstrating the complete causal chain from raw material characteristics and process parameters to the formation of flavor substances, providing a visual analysis tool for analyzing the complex causal relationships in the liquor brewing process.

[0032] S4: Based on the critical paths identified in the causal graph, a regression algorithm is used to obtain the path coefficients of the critical paths. The total effect from the raw material node to the flavor substance node through all direct and indirect paths is calculated using the path coefficients. The contribution of the raw material node is quantified based on the total effect. The associated path is the node path from the raw material node to the flavor substance node in the causal graph, which includes the raw material node, the influencing node, and the flavor substance node; the influencing node is any node on the node path other than the raw material node and the flavor substance node; the regression algorithm is the formula used to obtain the path coefficient; the path coefficient directly measures the strength and direction of the influence between variables, and the formula for obtaining the path coefficient is: ; ; in To quantify and standardize data that affects nodes; This is the baseline value for predicting the dependent variable when the independent variable is 0; it is also called the intercept term. for right Path coefficients; Quantitative and standardized data for raw material nodes; This represents the error term in the regression equation; Quantitative and standardized data for flavor compound nodes; for right Path coefficients; for right Path coefficients.

[0033] The total effect between the raw material node and the flavor substance node is calculated using the path coefficients. This total effect includes a first direct effect from the raw material node directly to the flavor substance node and a first indirect effect from the influencing node to the flavor substance node. The first indirect effect includes a second direct effect and a third direct effect. The second direct effect is the effect from the influencing node directly to the flavor substance node; the third direct effect is the effect from the raw material node directly to the influencing node. The formula for calculating the total effect is: ; ; ; in This is the first direct effect between the raw material node and the flavor substance node; This is the first indirect effect affecting the connection between nodes and flavor compound nodes; The total effect of raw material nodes reaching flavor substance nodes for right Path coefficients; for right Path coefficients; for right Path coefficients.

[0034] The contribution degree represents the degree of influence of raw material characteristics on flavor compounds. It is obtained by comparing the total effect of raw material nodes on flavor compounds with the total effect of all significant factors affecting that flavor compound. The formula for obtaining the contribution degree is as follows: ; in for right The degree of contribution; for right The overall effect; For all significant effects The sum of the total effects of the variables; The value is expressed as a percentage.

[0035] It should also be noted that, in analyzing the causal network of baijiu quality formation, a complete influence path from raw material characteristics to flavor substances was constructed. This path starts with the raw material node, proceeds through several intermediate variable nodes, and ultimately acts on the terminal flavor substance node, fully presenting the dynamic process of raw material components transforming into flavor characteristics.

[0036] To quantify the strength of causal interactions among variables in the path, a multiple regression model is used for parameter estimation. Firstly, the linear regression equation... Calculate raw material nodes For mediating variables Direct influence coefficient ;Including the quantitative and standardized data of raw material nodes This refers to the dimensionless value obtained after Z-Score standardization of the original detection index corresponding to the raw material node. The raw material sample is detected using a near-infrared spectroscopy instrument to obtain the original concentration value of the target component. The average detection value of this batch of raw materials is used as the initial quantification feature of this node. Subsequently, Z-Score standardization is performed on the initial quantification feature to obtain the standardized data. ; Quantitative and standardized data affecting nodes This refers to the dimensionless values ​​obtained after Z-Score standardization of the original detection indicators corresponding to the influencing node. The mean value representing the characteristics of each process stage is extracted from the real-time process parameter time series collected by a network of temperature, pH, and humidity sensors throughout the fermentation cycle, and used as the original quantitative indicator for that influencing node. Z-Score standardization is then performed on the original quantitative indicator to eliminate dimensional differences, resulting in the standardized data. Then, a complete model including direct and indirect paths is established: ,in Characterizing raw material nodes flavor compounds The direct effect intensity, This reflects the mediating variable. right The transmission effect; including the quantitative and standardized data of flavor substance nodes. Z-score standardization refers to the dimensionless value obtained after Z-score standardization of the original detection index corresponding to the flavor substance node. Ion current spectra of flavor substances are obtained using gas chromatography-mass spectrometry (GC-MS), and the characteristic peak area of ​​the target compound is calculated by integration. The peak area is then converted to the absolute concentration of the flavor substance in the wine sample using the internal standard curve method. Z-score standardization is then applied to the absolute concentration to obtain the standardized data. ; For the raw material node When the content is 0, the number of nodes in the basic sample that affect the nodes The content; To provide a baseline value for balancing errors caused by different circumstances, it is indirectly estimated by calculating the residual between the actual observed value and the model prediction value for each sample after model fitting; this is done when obtaining path coefficients. At that time, based on the standardized data of each node in the fusion database, and taking the dependent variable node of each directed edge in the causal graph as the target and the independent variable node as the input, respectively, construct... or The least squares method is used to estimate the parameters of each regression equation and obtain the regression coefficients of each independent variable, which are the path coefficients of the directed edge.

[0037] Based on the aforementioned path coefficients, the overall impact of raw materials on flavor compounds was further decomposed. Direct effects. From coefficients Direct measurement reflects the direct pathway by which raw materials affect flavor compounds; first indirect effect. Through the product term Calculate and characterize the cascade effects produced by raw materials through mediating variables. Total effect. This is the sum of the direct and indirect effects: This comprehensive index fully reflects the overall contribution of raw material components to flavor formation. When the mediating variable is not 1, the calculation of the first indirect effect requires treating the mediating variable adjacent to the raw material node as a new raw material node. After calculating all paths from the raw material to the flavor substance, the results are summed to obtain the total effect of the raw material on the flavor substance.

[0038] To facilitate cross-sectional comparisons of different raw material components, a standardized contribution index is established. This indicator is calculated by using specific raw materials. flavor compounds The percentage value is calculated as the proportion of the total effect of each ingredient to the sum of the total effects of all significant influencing factors, after normalization. This quantitative indicator not only clearly demonstrates the relative importance of each ingredient component in flavor formation, but also provides a data-driven decision-making basis for the precise optimization of ingredient ratios.

[0039] S5: Based on the metabolite information in the fusion database, select compounds that have been proven to have health benefits by literature or cell experiments, and calculate the health value index through beneficial metabolic data; After obtaining metabolite information from a fusion database, beneficial metabolites with health benefits confirmed by literature or cell experiments were screened. The standardized content of these beneficial metabolites was directly obtained from the fusion database. Then, the half-maximal inhibitory concentration (IC50) of each beneficial metabolite was determined using a DPPH free radical scavenging assay. Weights were assigned based on the ratio of the minimum IC50 value to the measured values ​​of different metabolites. A health value index was calculated using the standardized content and weighted distribution of different beneficial metabolites. The formula for obtaining the health value index is as follows: ; in As a health value index; For the first Standardized content of beneficial metabolites in wine samples; For the first The weight of beneficial metabolites; This is the summation symbol.

[0040] It should also be noted that in constructing the health value index, standardized metabolite profile data were first extracted from the fusion database, and beneficial compounds with clear physiological activities were screened based on existing literature research and in vitro experimental evidence. Subsequently, weighting coefficients were obtained based on the DPPH free radical scavenging experiment; the DPPH free radical scavenging experiment measures the DPPH free radical scavenging rate of the compound at different concentrations, and calculates its half-maximal inhibitory concentration, i.e., IC50 value, through dose-response curves. The lowest IC50 value among all tested beneficial metabolites was used as a benchmark, and different weighting coefficients were obtained by comparing the benchmark with the measured value of the tested compound, thereby accurately reflecting the relative importance of each component's contribution to health.

[0041] Finally, a weighted summation model was used. Calculate the overall score, where Indicates the first Standardized concentrations of various active ingredients This represents the experimentally validated weighting coefficient. The calculation model considers both the abundance level and biological potency of active ingredients, avoiding the one-sidedness of solely relying on concentration and overcoming the limitations of considering only activity intensity, thus achieving a scientific quantification of the health value of baijiu. This index system establishes a unified standard for the objective comparison of product health attributes, helping producers optimize processes and providing consumers with a scientific reference for their choices.

[0042] S6: Generate a visual report based on the sensory flavor score, causal graph, contribution, and health value index.

[0043] The visualization report includes: directly displaying the product's performance across multiple sensory dimensions in the form of a radar chart using the sensory flavor score; showing the main pathways affecting the product's key flavors using the cause-and-effect diagram; displaying the product's main ingredients and their content using the contribution index, and listing the top-ranking ingredients by contribution and their contribution values; and displaying the HVI score and level in the form of a dashboard using the health value index.

[0044] It should also be noted that the evaluation report generated by this system adopts a multi-dimensional visualization scheme to comprehensively present the overall evaluation results of the liquor quality. In terms of sensory characteristic display, a radar chart is used to construct a flavor profile model, clearly showing the score distribution of the product in key dimensions such as aroma complexity, mouthfeel fullness, and aftertaste persistence. This helps users quickly establish an overall understanding of the product's sensory characteristics. The quantitative representation of the flavor profile model is achieved by averaging the standardized scores of multiple tasters on each sensory dimension, and then extracting the principal component scores representing the core sensory characteristics after dimensionality reduction through principal component analysis.

[0045] To provide a deeper understanding of flavor formation mechanisms, the report introduces a causal network graph, visually representing the complete causal path from raw material characteristics to process parameters, ultimately influencing key flavor compounds, through the topological structure of nodes and directed edges. This visualization not only reveals the intrinsic connections between various elements but also provides a theoretical basis for the precise control of production processes.

[0046] In the raw material analysis section, the report uses a contribution ranking table to quantitatively display the contribution weight of various raw material components to flavor substances and their actual content. This ranking result is based on a feature importance algorithm, which helps producers accurately identify key raw material elements that affect product flavor.

[0047] In addition, the report integrates a dynamic monitoring dashboard for the health value index. This dashboard displays the product's health value score and corresponding grade range in real time through pointer position and color gamut changes, providing a clear visual representation of the product's health attributes. This multi-chart collaborative visualization architecture ensures both accurate transmission of professional data and easy readability of the results, providing decision-makers with a comprehensive perspective on quality assessment.

[0048] Example 2: Based on Example 1, a liquor quality evaluation system based on multi-source data analysis, such as... Figure 2 As shown, it includes: The data acquisition module is responsible for systematically collecting four major categories of data to provide basic data for subsequent data processing and analysis. The data processing module cleans, aligns, and standardizes the collected multi-source data to improve data quality and eliminate the impact of different units and orders of magnitude on the model. The feature engineering module extracts multidimensional features from raw data and constructs a feature matrix for causal discovery and machine learning modeling. The data fusion module creates a core database that links data from all dimensions together to form a structured wide table for causal discovery and machine learning modeling. The reasoning and calculation module uses a causal discovery algorithm to generate a causal graph, quantifies the contribution of each feature, and calculates the health value index, providing a basis for the quality analysis and optimization of baijiu. The evaluation output module integrates scores from multiple dimensions into a comprehensive index and generates a visual report, providing an intuitive reference for the production and decision-making of baijiu.

[0049] Although the present invention has been described in detail with reference to the above embodiments, it will be apparent to those skilled in the art that various changes or modifications can be made to the invention without departing from the principles and spirit of the invention as defined by the claims. Therefore, the detailed description of the embodiments in this disclosure is for illustrative purposes only and is not intended to limit the invention; rather, the scope of protection is defined by the content of the claims.

Claims

1. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis, characterized in that: Includes the following steps: S1: Obtain multi-source data information of liquor based on various detection and analysis methods, and perform preprocessing operations on the multi-source data information, which includes raw material data, process data, product data and external data; S2: Multidimensional feature extraction is performed based on the processed product data, and a fusion database is constructed using the extracted multidimensional feature data and processed multi-source data information. The fusion database includes wine sample information, raw material feature information, process feature information, flavor substance information, metabolite information, and sensory flavor score. S3: Based on the raw material characteristic information, process characteristic information and flavor substance information in the constructed fusion database, a causal graph is generated by executing a causal discovery algorithm. The causal graph is a directed acyclic graph. S4: Based on the critical paths identified in the causal graph, a regression algorithm is used to obtain the path coefficients of the critical paths. The total effect from the raw material node to the flavor substance node through all direct and indirect paths is calculated using the path coefficients. The contribution of the raw material node is quantified based on the total effect. S5: Based on the metabolite information in the fusion database, select compounds that have been proven to have health benefits by literature or cell experiments, and calculate the health value index through beneficial metabolic data; S6: Generate a visual report based on the sensory flavor score, causal graph, contribution, and health value index.

2. The method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 1, characterized in that, The process involves acquiring multi-source data information on baijiu (Chinese liquor) using various detection and analysis methods, and preprocessing this data. The multi-source data includes raw material data, process data, product data, and external data. Specifically, the multi-source data acquired using these methods includes: raw material data obtained through near-infrared spectroscopy, high-performance liquid chromatography, ultraviolet spectrophotometry, and biochemical analyzers; process data obtained through temperature sensors, humidity sensors, pH sensors, and data acquisition devices; flavor substance data obtained through gas chromatography-mass spectrometry; metabolic data obtained through liquid chromatography-high-resolution mass spectrometry; sensory evaluation data obtained through a tasting panel and standardized scoring sheets; and data obtained through market research. External data obtained through web crawling and sales data analysis; the preprocessing operations include handling missing values, removing outliers, time series alignment, and data standardization; handling missing values ​​refers to missing data caused by accidental sensor malfunction or detection failure, which is filled by interpolation or based on other samples from the same batch; removing outliers refers to identifying and removing abnormal detection values ​​that significantly deviate from the normal range using statistical methods; time series alignment refers to accurately mapping the time series data of process parameters to specific brewing batches; data standardization refers to processing continuous numerical data using Z-Score standardization and digitizing descriptive terms for sensory evaluation; the product data includes flavor substance data, metabolic data, and sensory evaluation data.

3. The method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 2, characterized in that, The process involves multidimensional feature extraction based on processed product data, and the construction of a fusion database using the extracted multidimensional feature data and processed multi-source data. This fusion database includes wine sample information, raw material characteristic information, process characteristic information, flavor substance information, metabolite information, and sensory flavor scores. Specifically, the multidimensional feature extraction based on processed product data includes extracting the absolute concentration or relative peak area of ​​specific flavor compounds as features, summarizing flavor substance information based on these features, and comparing the metabolic data with the database after peak identification, peak alignment, and noise reduction using software to qualitatively identify some compounds. A metabolic feature matrix containing the peak intensities of all samples and all detected compounds is generated, and metabolite information is summarized based on the metabolic feature matrix. By averaging the scores from multiple tasters and performing principal component analysis for dimensionality reduction, the principal component scores representing the core sensory dimensions are extracted as features, and sensory flavor scores are summarized based on these features. The construction of a fusion database using the extracted multidimensional feature data and processed multi-source data involves associating the flavor substance information, metabolite information, sensory flavor scores, and raw material and process feature information from the multi-source data through wine sample information, ultimately constructing a fusion database.

4. The method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 3, characterized in that, The raw material feature information, process feature information, and flavor substance information in the constructed fusion database are used to generate a causal graph by executing a causal discovery algorithm. The causal graph is a directed acyclic graph (DAG). The causal discovery algorithm is an algorithm that infers the causal direction between variables based on assumptions and statistical principles. The main causal discovery algorithm used is the PC algorithm. The PC algorithm starts from a completely undirected graph based on conditional independence tests, and gradually deletes edges through conditional independence tests to determine the undirected association between raw material features, process features, and flavor substance information. Then, it assigns directionality to the remaining connecting edges according to the causal direction determination rules, and finally obtains a directed acyclic graph to represent the causal relationship between raw material features, process features, and flavor substance information, and to show the causal influence path to form a causal graph. The nodes in the directed acyclic graph are each feature or flavor substance, and the edges represent causal relationships.

5. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 4, characterized in that, The step of obtaining path coefficients for critical paths identified in the causal graph using a regression algorithm includes: the associated path being the node path from the raw material node to the flavor substance node in the causal graph, which includes the raw material node, the influencing node, and the flavor substance node; the influencing node being other nodes on the node path besides the raw material node and the flavor substance node; the regression algorithm being the formula used to obtain the path coefficients; and the path coefficients directly measuring the strength and direction of the influence between variables, wherein the formula for obtaining the path coefficients is: ； ； in To quantify and standardize data that affects nodes; This is the baseline value for predicting the dependent variable when the independent variable is 0; it is also called the intercept term. for right Path coefficients; Quantitative and standardized data for raw material nodes; This represents the error term in the regression equation; Quantitative and standardized data for flavor compound nodes; for right Path coefficients; for right Path coefficients.

6. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 4, characterized in that, The calculation of the total effect from the raw material node to the flavor substance node via all direct and indirect paths using the path coefficients includes: calculating the total effect between the raw material node and the flavor substance node using the path coefficients; the total effect includes a first direct effect between the raw material node and the flavor substance node and a first indirect effect between the influencing node and the flavor substance node, the first indirect effect including a second direct effect and a third direct effect, the second direct effect being the effect between the influencing node and the flavor substance node; the third direct effect being the effect between the raw material node and the influencing node, wherein the formula for calculating the total effect is: ； ； ； in This is the first direct effect between the raw material node and the flavor substance node; This is the first indirect effect affecting the connection between nodes and flavor compound nodes; The total effect of raw material nodes reaching flavor substance nodes for right Path coefficients; for right Path coefficients; for right Path coefficients.

7. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 4, characterized in that, The step of quantifying the contribution of the raw material node based on the total effect includes: the contribution represents the degree of influence of the raw material characteristics on the flavor substance, and is obtained by comparing the total effect of the raw material node on the flavor substance with the total effect of all significant influences on the flavor substance, wherein the formula for obtaining the contribution is: ； in for right The degree of contribution; for right The overall effect; For all significant effects The sum of the total effects of the variables; The value is expressed as a percentage.

8. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 5, characterized in that, The process of screening compounds with proven health benefits from literature or cell experiments based on metabolite information in the fusion database, and calculating a health value index using beneficial metabolic data, includes: obtaining metabolite information from the fusion database and screening for beneficial metabolites with proven health benefits from literature or cell experiments; directly obtaining the standardized content of the beneficial metabolites from the fusion database; determining the half-maximal inhibitory concentration (IC50) of each beneficial metabolite using a DPPH free radical scavenging assay; obtaining weights based on the ratio of the minimum IC50 value to the measured values ​​of different metabolites; and calculating the health value index using the standardized content of different beneficial metabolites and the weight allocation. The formula for obtaining the health value index is as follows: ； in As a health value index; For the first Standardized content of beneficial metabolites in wine samples; For the first The weight of beneficial metabolites; This is the summation symbol.

9. A method for evaluating the quality of baijiu (Chinese liquor) based on multi-source data analysis as described in claim 6, characterized in that, The process of generating a visualization report based on the sensory flavor score, cause-effect diagram, contribution rate, and health value index includes: the visualization report directly displaying the product's performance across multiple sensory dimensions in the form of a radar chart using the sensory flavor score; displaying the main pathways affecting the product's key flavors using the cause-effect diagram; displaying the product's main components and their content using the contribution rate, and listing the top-ranking components by contribution rate and their contribution values; and displaying the HVI score and level using the health value index in the form of a dashboard.

10. A liquor quality evaluation system based on multi-source data analysis, used to implement the liquor quality evaluation method based on multi-source data analysis described in any one of 1-9, characterized in that, include: The data acquisition module is responsible for systematically collecting four major categories of data to provide basic data for subsequent data processing and analysis. The data processing module cleans, aligns, and standardizes the collected multi-source data to improve data quality and eliminate the impact of different units and orders of magnitude on the model. The feature engineering module extracts multidimensional features from the raw data and constructs a feature matrix for subsequent causal discovery and machine learning modeling. The data fusion module creates a core database that links data from all dimensions together to form a structured wide table for subsequent causal discovery and machine learning modeling. The reasoning and calculation module uses a causal discovery algorithm to generate a causal graph, quantifies the contribution of each feature, and calculates the health value index, providing a basis for the quality analysis and optimization of baijiu. The evaluation output module integrates scores from multiple dimensions into a comprehensive index and generates a visual report, providing an intuitive reference for the production and decision-making of baijiu.

Images