A traditional Chinese medicine extraction quality management method based on data analysis

Abstract

This invention discloses a data analysis-based method for quality management of traditional Chinese medicine (TCM) extraction, belonging to the field of TCM pharmaceutical process control and intelligent manufacturing technology. The method identifies key control points throughout the extraction and concentration process, formulates a key process parameter acquisition plan, collects multi-source heterogeneous production data, preprocesses it, establishes a standard curve of solid content variation with production time, constructs a predictive model after completing key parameter correlation analysis, and clarifies the solid content control threshold. Subsequently, it collects production process data in real time. When real-time monitoring data or predictive results exceed the control threshold, multiple process adjustment schemes are generated, and the execution effect of process adjustments is recorded to achieve continuous process optimization. This invention can significantly improve the stability and batch quality consistency of the TCM extraction process, transforming the quality control model from post-process control to pre-process prediction and in-process intervention, fully leveraging the value of production data, and meeting the requirements of GMP compliance and quality process control.

Description

Technical Field

[0001] This invention relates to a quality management method, specifically a data analysis-based method for quality management of traditional Chinese medicine extraction, belonging to the field of traditional Chinese medicine manufacturing process control and intelligent manufacturing technology. Background Technology

[0002] Extraction of Chinese medicinal herbs is a core step in the production of traditional Chinese medicine (TCM). It is a crucial process for separating and enriching the effective active ingredients from raw materials. The extraction process directly determines the retention rate of active ingredients in the herbs, thus affecting the efficacy, stability, and clinical effectiveness of the TCM product. With the rapid development of the TCM industry, the market demands increasingly higher batch-to-batch consistency in TCM products. Fluctuations in the content of active ingredients between different batches not only affect product efficacy but also reduce the consistency of user experience, hindering market acceptance and industrial promotion of TCM products.

[0003] Currently, the quality control model in the traditional Chinese medicine extraction industry still relies primarily on traditional process sampling and final product inspection, a typical post-production control model. Under this model, online process monitoring points are generally insufficient, making it impossible to monitor process parameters and quality indicators throughout the entire extraction process in real time and continuously. Quality test results are usually only available during production or after a batch is completed, making it difficult to predict and intervene in quality anomalies. Once process parameters deviate or quality indicators fail to meet standards, the substandard product has already been produced, resulting in significant waste of raw materials and energy, increasing production costs, failing to meet the quality control requirements of drug regulatory authorities, and failing to guarantee consistent user experience across different batches. Furthermore, the traditional Chinese medicine extraction process generates massive amounts of process data, including equipment operating parameters, production environment monitoring data, and online testing data. However, currently, most of this data is scattered across different production equipment, control systems, and management platforms, lacking effective data exchange and integration. These massive amounts of data are only used for simple process monitoring and historical records. The technological patterns and quality correlation information contained in them have not been fully explored and utilized. They cannot be used to guide process optimization and quality improvement through in-depth correlation analysis of the data. As a result, the control of the Chinese medicine extraction process has long relied on the personal experience of the operators, making it difficult to achieve precise and stable quality control.

[0004] In response to the national strategy for the modernization and intelligentization of traditional Chinese medicine, there is an urgent need in this field to develop a quality management method for the extraction of traditional Chinese medicine that can break through the limitations of the traditional post-production quality control model, realize the integrated management and in-depth analysis of production process data, improve the quality process management capability, and ultimately ensure the stability of the content of product efficacy components and the consistency of user experience. Summary of the Invention

[0005] To overcome the aforementioned shortcomings of existing technologies, this invention proposes a data analysis-based method for quality management of traditional Chinese medicine extraction, enabling pre-emptive prediction and in-process intervention of quality anomalies, thereby improving the stability of solid content and batch-to-batch quality consistency. The technical solution adopted by this invention is as follows: A data analysis-based method for quality management of traditional Chinese medicine extraction includes the following steps: S1: Analyze the entire process of traditional Chinese medicine extraction and concentration, identify key quality control points that affect solid content, and formulate a key process parameter collection plan; S2: Collect multi-source heterogeneous production data, preprocess the collected data, and obtain a high-quality training dataset; S3: Based on the high-quality training dataset described in S2, establish a standard curve of solid content changing with production time, and construct a prediction model after completing the correlation analysis of key parameters. S4: Train the prediction model constructed in S3 and initialize the system model library, and define the solid content control threshold; S5: Real-time acquisition of production process data, and extrapolation of the changing trend of solid content based on the standard curve established in S3 and the prediction model trained in S4; S6: When real-time monitoring data or prediction results exceed the control threshold, multiple process adjustment schemes are generated; S7: Record the execution effect of process adjustments, write the data back to the model library, optimize the prediction model and adjustment strategy, and achieve continuous process optimization.

[0006] Furthermore, the multi-source heterogeneous production data mentioned in S2 includes the original process parameter data of the DCS system, the supplementary process parameter data collected by the newly added sensors, and the real-time data of solid content and active ingredients collected by the online detection equipment.

[0007] Furthermore, the preprocessing adopts a three-level cleaning method, which involves sequentially removing outliers, filling in missing values, and replacing duplicate values.

[0008] Furthermore, the method for establishing the standard curve of solid content change with production time is as follows: S31: At each time point, aggregate no less than 30 batches of historical production data horizontally, and calculate the mean and 95.4% confidence interval of the data at each time point; S32: The interval boundaries are smoothed using cubic spline method or smooth spline method to obtain a preliminary curve sequence; S33: Construct a multiple linear regression model, use the least squares method to solve for the polynomial parameters, and combine the pharmacopoeia standards and process specifications to form hard constraint boundaries; S34: Perform integral verification on the area under the standard curve, evaluate the model's coefficient of determination and root mean square error, and iteratively optimize to obtain the final solids content and time standard curve.

[0009] Furthermore, the key parameter correlation analysis specifically involves: using Pearson correlation coefficient, cosine similarity, and feature ablation test algorithms to calculate and classify the correlation strength between each key process parameter and solid content; using principal component analysis to reduce the dimensionality of high-dimensional data; and using dynamic time warping algorithm to align time-lag data.

[0010] Furthermore, the solid content control threshold is ±10%.

[0011] Furthermore, the prediction model adopts a multi-stage architecture: when the sample size is less than a preset threshold, a differential autoregressive moving average model is used for short-term prediction; when the sample size reaches the preset threshold, the model is switched to a long short-term memory network model for multivariate time-series prediction.

[0012] Furthermore, based on the fluctuation range of the solid content standard curve and the prediction model, a dual-threshold alarm mechanism with high and low thresholds is constructed.

[0013] Furthermore, the process adjustment scheme includes key parameter adjustment values, predicted improvement range of solid content, and confidence level, which are obtained by constructing and solving a multi-objective optimization constraint set that includes the physical range of process parameters, adjustment costs, and solid content control requirements.

[0014] A data analysis-based quality management system for traditional Chinese medicine extraction, wherein the system operates using the aforementioned data analysis-based quality management method for traditional Chinese medicine extraction, comprising: The key process parameter centralized control platform is used to realize data acquisition, real-time monitoring, anomaly early warning and historical traceability of various key process parameters in the production process; The solids content modeling and analysis system is used to establish a mechanistic correlation model between solids content and time and process parameters, so as to achieve the prediction, evaluation and optimization control of quality indicators.

[0015] The beneficial effects of this invention are: 1. This invention transforms the traditional passive quality control model of post-production sampling and finished product inspection into a proactive management model of pre-production prediction and in-process intervention. By using a multi-stage predictive model to anticipate the trend of solid content changes and combining it with a high and low dual-threshold alarm mechanism, quality risks can be identified in advance and targeted adjustment plans can be generated, significantly reducing the generation of batches of unqualified products and reducing the waste of raw materials and production energy in traditional Chinese medicine.

[0016] 2. This invention integrates multi-source heterogeneous production data scattered across different devices and systems through unified data collection, standardized processing, and integrated storage. It establishes a quantitative correlation between process parameters and quality indicators, transforming dormant production data into knowledge that can guide process control, and promoting the transformation of production management from experience-driven to data-driven.

[0017] 3. By establishing a standard curve of solid content change over time and combining it with a multi-objective optimization adjustment strategy, the present invention strictly controls the fluctuation of solid content within ±10%, effectively ensuring the stability of the retention rate of active ingredients in different batches of products, thereby improving the consistency of product efficacy and user experience.

[0018] 4. This invention continuously optimizes the prediction model and adjustment strategy through incremental training, which can adapt to process changes such as different batches of raw materials and equipment aging, and achieve continuous optimization and long-term stable operation of the production process. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating one implementation of the data analysis-based quality management method for traditional Chinese medicine extraction according to the present invention. Figure 2 This is a schematic diagram of one embodiment of the traditional Chinese medicine extraction quality management system based on data analysis of the present invention; Figure 3 This is a schematic diagram of one implementation of the correlation analysis between key parameters and solid content of the present invention. Detailed Implementation

[0020] Specific implementation method one: Combining Figure 1-3 This implementation method is described as follows: Figure 1 As shown in this embodiment, a data analysis-based method for quality management of traditional Chinese medicine extraction includes the following steps: S1: Analyze the entire process of traditional Chinese medicine extraction and concentration, identify key quality control points that affect the solid content (Y); assess the online monitoring capabilities of existing equipment, formulate a key process parameter (X) acquisition plan, and achieve real-time acquisition, transmission and dynamic control.

[0021] S2: Collect multi-source heterogeneous production data, preprocess the collected data to obtain a high-quality training dataset. The multi-source heterogeneous production data includes the original process parameter data of the DCS control system, the supplementary process parameter data collected by newly added sensors, and the real-time data of solid content and active ingredients collected by online detection equipment.

[0022] The supplementary process parameter data collected by the newly added sensors include: temperature in the middle of the extraction tank, temperature difference between the upper and lower parts of the extraction tank, pressure in the outlet pipe of the extraction tank, and circulation flow rate of the extraction tank. The data collected by the online detection equipment includes: real-time solid content data collected by an online refractometer installed at the top of the filter in the lower drug outlet pipe of the extraction tank; and spectral data of the drug solution collected by a near-infrared spectrometer installed on the main extraction circulation pipe. Based on a pre-established quantitative correction model (which converts online spectral data into solid content and active ingredient content data), real-time concentration data of various active ingredients are calculated. The method for establishing the quantitative correction model for the near-infrared spectrometer is as follows: simultaneously collecting a large amount of online spectral data and corresponding offline HPLC analysis data, and using a partial least squares (PLS) algorithm to establish a quantitative relationship between the spectral data and the content of active ingredients.

[0023] The preprocessing employs a three-stage cleaning method, which involves sequentially removing outliers, filling in missing values, and replacing duplicate values. The outlier removal process uses a 3-stage cleaning method. The criteria and box plot combined judgment method remove erroneous data caused by sensor jumps and sampling errors; the missing value completion method uses linear interpolation to complete data with 5 or fewer consecutive missing values, and uses K-nearest neighbor interpolation method of the same batch of data to complete data with long-term missing values; the duplicate value replacement method retains the first record or replaces duplicate data with the mean value at that time.

[0024] S3: Based on the high-quality training dataset described in S2, establish a standard curve of solid content changing with production time, and construct a prediction model after completing the correlation analysis of key parameters.

[0025] The method for establishing the standard curve of solid content change with production time is as follows: S31: At each time point, aggregate no less than 30 batches of historical production data horizontally, and calculate the mean and 95.4% confidence interval of the data at each time point; S32: The interval boundaries are smoothed using cubic spline method or smooth spline method to obtain a preliminary curve sequence; S33: Construct a multiple linear regression model, use the least squares method to solve for the polynomial parameters, and combine the pharmacopoeia standards and process specifications to form hard constraint boundaries; S34: Integral verification is performed on the area under the standard curve to evaluate the model's coefficient of determination and root mean square error. Iterative optimization yields the final solids content and time standard curve. The generated standard curve is stored in the database as a function expression and parameter set, and supports dynamic calling and visualization by medicinal material variety, production batch, and process type.

[0026] The key parameter correlation analysis specifically involves using Pearson correlation coefficient, cosine similarity, and feature ablation test algorithms to calculate and classify the correlation strength between each key process parameter and solid content. Specifically, the correlation coefficients are divided into five levels, forming a parameter correlation combination database. Principal component analysis (PCA) is used for dimensionality reduction of high-dimensional data, and dynamic time warping (DTW) is used for alignment of time-lag data.

[0027] The prediction model employs a multi-stage architecture: when the sample size is less than a preset threshold, a differential autoregressive moving average model is used for short-term prediction; once the sample size reaches the preset threshold, the model switches to a long short-term memory network model for multivariate time-series prediction. This model can quickly output predicted values ​​and confidence intervals for several future steps, providing a reference for early intervention in the production process.

[0028] S4: Train the prediction model constructed in S3 and initialize the system model library, and clarify the solid content control threshold; preferably, the solid content control threshold is ±10%. Controlling the solid content fluctuation within ±10% can effectively ensure the stability of the traditional Chinese medicine extraction process and the batch-to-batch quality uniformity, ensuring stable content of effective components and consistent efficacy of the product; it can also avoid excessively frequent process adjustments and a significant increase in production difficulty and cost due to excessively high control precision requirements, achieving an optimal balance between quality stability and production operability, while meeting the requirements of pharmacopoeia standards, process specifications, and GMP for the controllability of traditional Chinese medicine product quality.

[0029] By establishing a standard curve of solid content change over time and combining it with a multi-objective optimization and adjustment strategy, the fluctuation of solid content is strictly controlled within ±10%, which effectively ensures the stability of the retention rate of active ingredients in different batches of products, thereby improving the consistency of product efficacy and user experience.

[0030] S5: Real-time acquisition of production process data; based on the standard curve established in S3 and the prediction model trained in S4, the trend of solid content changes is inferred. Key process parameters and solid content data are acquired in real-time during production, and pre-trained prediction models from the model library are used to infer future trends in solid content changes. Preferably, a dual-threshold alarm mechanism is constructed based on the fluctuation range of the solid content standard curve and the prediction model. When data exceeds the threshold range, local anomaly trend analysis is automatically triggered and a global anomaly warning message is generated.

[0031] S6: When real-time monitoring data or prediction results exceed the control threshold, multiple process adjustment schemes are generated for operators to choose from. The process adjustment scheme includes key parameter adjustment values, the predicted increase in solid content, and confidence level. It is obtained by constructing and solving a multi-objective optimization constraint set that includes the physical range of process parameters, adjustment costs, and solid content control requirements. The method for generating the process adjustment scheme is as follows: using the current environment and process parameters as initial conditions, the future trend of solid content changes is extrapolated to determine the time nodes for intervention initiation, termination, and recovery; a multi-objective optimization constraint set that includes the physical range of process parameters, adjustment costs, and solid content control requirements is constructed; the local hyperparameter optimal solution is solved to generate multiple process adjustment schemes that include key parameter adjustment values, the predicted increase in solid content, and confidence level.

[0032] S7: Record the execution effect of process adjustments, write the data back to the model library, optimize the prediction model and adjustment strategy, and achieve continuous process optimization. Incrementally self-train the prediction model and optimize the adjustment strategy to form a closed-loop quality management system encompassing data collection, monitoring, analysis, early warning, adjustment, and optimization. This enables continuous optimization and long-term stable operation of the production process.

[0033] like Figure 2 As shown, a data analysis-based quality management system for traditional Chinese medicine extraction is disclosed. The system operates using the aforementioned data analysis-based quality management method for traditional Chinese medicine extraction, and includes: The key process parameter centralized control platform is used to realize data acquisition, real-time monitoring, anomaly early warning and historical traceability of various key process parameters in the production process; The solids content modeling and analysis system is used to establish a mechanistic correlation model between solids content and time and process parameters, so as to achieve the prediction, evaluation and optimization control of quality indicators.

[0034] The data analysis-based quality management system for traditional Chinese medicine extraction includes a data source layer, a data acquisition layer, a data storage layer, a data modeling and analysis layer, and an application service layer that are connected in sequence via communication. Data source: This layer manages communication between existing industrial control systems and databases, and uses standard protocols, remote database access, and interface calls to achieve platform integration; Data Acquisition: This layer is divided into real-time acquisition and batch synchronization. The former is the real-time perception and data acquisition of the monitored objects, and fast real-time read and write is achieved through a caching mechanism; the latter is to achieve batch synchronization of low-frequency historical data through a remote database to support functions such as data persistence, data cleaning, data backup and recovery. Data storage: This layer achieves high-frequency read / write performance and data security consistency of the database through caching and data persistence technologies. Caching is used to write frequently collected data, model parameters, and weights into a buffer, enabling fast data retrieval and efficient rule usage. Simultaneously, this data is incrementally backed up to the main database at regular intervals, supporting offline model updates and optimization, data disaster recovery, and archiving cold data (such as data older than 3 years) to reduce online storage costs. Data Modeling and Analysis: This layer incorporates a system feature engine to manage training features for operator-generated models, such as sliding window statistics, DTW distance, Pearson correlation coefficient, cosine similarity, and PCA dimensionality reduction. This layer also manages the system's model repository, including traditional-deep model pools such as ARIMA, LSTM, association analysis, and weighted feedback, to provide system-related analytical services. Application Services: This layer provides visualization services for system functions, transforming complex model outputs into understandable, easy-to-use, and manageable business actions. It supports functions such as viewing standard curves, online monitoring of key parameters, solid content monitoring and control schemes, parameter correlation query and dynamic visualization, system abnormal operating condition detection and early warning, and basic system management, all centered around monitoring, early warning, execution, and feedback.

[0035] Specifically, the data source layer is used to establish communication with the industrial control system and database to obtain production data during the extraction process of traditional Chinese medicine. The production data includes production equipment interface data and real-time sensor data. The production equipment interface data is PLC system data of the extraction tank and concentration tank, including extraction temperature, extraction time, extraction pressure, stirring rate, concentration temperature, concentration time and material-liquid ratio. The real-time sensor data includes solvent usage and material flow rate collected by temperature sensors, pressure sensors and flow sensors.

[0036] The data acquisition layer is used to collect production data in real time. It includes a real-time acquisition module and a batch synchronization module. The real-time acquisition module is used to detect and collect production data from the traditional Chinese medicine extraction process in real time. The batch synchronization module is used to synchronize historical production data in batches. The data acquisition layer also includes a real-time database. Both the real-time acquisition module and the batch synchronization module are communicatively connected to the real-time database. The real-time acquisition module transmits the collected real-time data from the traditional Chinese medicine extraction process to the real-time database, and the batch synchronization module transmits historical production data synchronized from an external remote database to the real-time database. The real-time acquisition module employs a caching mechanism to temporarily cache and quickly read / write the high-frequency real-time data from the collected traditional Chinese medicine extraction process. The data acquisition layer collects production data in real time in the following ways: For production sections equipped with a DCS system, the real-time collected production data is transmitted to the real-time database through the OPC and DDE standard data communication interfaces provided by the DCS system, after being encapsulated by interface software; for production sites equipped with conventional simulation instruments, production data is collected through remote terminal devices and communicates with the real-time database via industrial Ethernet or fieldbus.

[0037] The data storage layer is used to classify and store the production data collected by the data acquisition layer, and to construct a standardized conceptual model and logical model for the data. It includes an internal database storage module and an external database storage module. The internal database storage module is used to store real-time process data of traditional Chinese medicine extraction and production, and the external database storage module is used to store operational monitoring data of traditional Chinese medicine extraction and production. The internal database storage module is implemented based on the built-in storage unit of WinCC software, and the external database storage module is built based on a relational database, which is a specific application unit of the relational database in this system.

[0038] The data modeling and analysis layer is a data analysis platform built on the C# high-level programming language. It is used to perform calculations, processing, and analysis on the production data stored in the data storage layer, generating analysis results. It includes a report management module, a data analysis module, an energy consumption analysis module, and an extraction rate analysis module. The report management module is used for centralized management and querying of reports in the production process; The data analysis module is used to retrieve core parameter data of the production process, realize the curve display and calculation analysis of the parameters, output the extreme values, average values ​​and rate of change of the parameters, and analyze the degree of influence of process parameters on the extraction rate of traditional Chinese medicine.

[0039] The energy consumption analysis module is used to analyze and calculate the operational monitoring data during the extraction of Chinese medicinal materials to obtain the energy consumption data of the production process. The extraction rate analysis module includes an extraction rate prediction unit and an actual extraction rate calculation unit. The extraction rate prediction unit calculates and outputs a predicted extraction rate value based on a mathematical model corresponding to key process parameters and the extraction rate, supports the optimization of the extraction rate prediction model, and analyzes the influence of key process parameters on the extraction rate. The actual extraction rate calculation unit calculates the actual extraction rate of the traditional Chinese medicine based on production data. This module provides extraction rate prediction and actual extraction rate calculation functions, optimizes the extraction rate prediction model, and analyzes the influence of key process parameters on the extraction rate.

[0040] The application service layer is a monitoring and management system, which includes a monitoring station, an operator station, a historical curve query module, a reporting system, an alarm information management module, a user management module, and a report printing module. This system enables remote operation, real-time monitoring, user management, and historical data traceability of the traditional Chinese medicine extraction and production process. It also provides visualized monitoring, parameter query, status alerts, and system management functions based on the analysis results from the data modeling and analysis layer.

[0041] The data-driven TCM extraction quality management system collects, stores, analyzes, and manages data, enabling intelligent and refined quality management of the TCM extraction process. It achieves real-time data collection, integrated analysis of the entire process, intelligent early warning of quality risks, and full traceability, thus improving the stability, uniformity, and controllability of TCM decoction pieces quality.

[0042] Example S1: Analyze the entire process of traditional Chinese medicine extraction and concentration, identify key quality control points that affect the solid content (Y); assess the online monitoring capabilities of existing equipment, formulate a key process parameter (X) acquisition plan, and achieve real-time acquisition, transmission and dynamic control.

[0043] S2: Collect heterogeneous production data from multiple sources, preprocess the collected data, and obtain a high-quality training dataset.

[0044] Specifically, data can be collected in the following ways: The control system increases the temperature point in the middle of the extraction tank and combines it with the temperature at the top of the extraction tank to detect the temperature deviation inside the extraction tank in real time. Combined with the control parameters of the extraction tank (cycle time, cycle interval time), it monitors the impact of the temperature difference inside the extraction tank on the proportion of solids content in real time.

[0045] The control system adds a pressure point to the liquid outlet pipeline of the extraction tank to monitor the impact of pipeline blockage on the content of extracted solids in real time.

[0046] The control system adds a concentration point (refractometer, concentration meter) at the outlet of the extraction tank, which is installed on the filter inlet to detect the content of extracted solids in real time. It also establishes a real-time curve of solid content (Y) changing with production time, with time as the axis, to provide a data basis for subsequent analysis.

[0047] The control system has been upgraded with a new extraction tank spectral analysis system (near-infrared spectrometer), which is installed on the circulation pipeline to collect spectra in real time and build a model to instantly calculate the real-time concentration of multiple active ingredients / components in the medicinal solution. This enables a rapid and non-destructive conversion from spectrum to content, generating a dynamic curve of time-multiple active ingredient content.

[0048] The control system combines the new and original points on the extraction tank to detect the solid content (Y) in real time and collect various data (X) that may affect the solid content, such as the upper and lower temperatures of the extraction tank, the jacket pressure of the extraction tank, the decoction time, and the decoction pressure setting, so as to provide a solid data foundation for subsequent analysis.

[0049] To obtain a clean, complete, and consistent high-quality data foundation for establishing high-precision standard curves, the data analysis-based TCM extraction quality management system unifies heterogeneous data from existing systems, instruments, and meters into a database. It aligns historical production processes according to the material-batch-process-timestamp hierarchy, forming a traceable raw dataset. Based on this, a three-level cleaning process is performed on the raw data, including: (1) Outlier removal, using 3 The criteria and box plots are used together to effectively obtain non-standard data points under normal distribution, thereby removing erroneous data such as sensor jumps and sampling errors. (2) Missing value completion: For data missing in a short period of time (five or fewer consecutive data values ​​missing), linear interpolation is used to fill in the missing values. This is an efficient completion method for short-term stable sequences. It can estimate missing values ​​using the mean, median, and polynomial. For data missing over a long period of time, the K nearest neighbor interpolation method of the same batch of data is used. By finding the K data that are most similar to the target data, the missing values ​​are estimated and predicted, thus maintaining the continuity of the time series. (3) Replace duplicate values, retain the first record or replace it with the mean at that time, to avoid sample weight distortion when building the standard curve.

[0050] S3: Based on the high-quality training dataset described in S2, establish a standard curve of solid content changing with production time, and construct a prediction model after completing the correlation analysis of key parameters.

[0051] Key parameter correlation analysis: To achieve quantitative correlation and precise process control between process parameters and solid content (Y) under existing production conditions, a key parameter correlation analysis module was constructed within the data analysis-based traditional Chinese medicine extraction quality management system. This module, based on the volume, feature dimensions, stationarity, and time-series lag of the actual production data, employs three algorithms—Pearson correlation coefficient, cosine similarity, and feature ablation test—to calculate and compare the correlation strength between each variable and Y, as well as among the variables themselves. Different processing strategies were designed for different data characteristics: For high-dimensional and complex correlation scenarios, principal component analysis (PCA) is used to linearly combine and reduce the dimensionality of the original features, generating clean and independent new variable combinations, thereby improving the accuracy and interpretability of correlation analysis.

[0052] To address the issue of misalignment between control and response times in sensor data from different process positions, a dynamic time warping algorithm is introduced to precisely align the lag sequences, significantly improving the reliability of time-related calculation results.

[0053] Based on the above, the system categorizes the correlation between various process parameters and Y into five levels according to the magnitude of the correlation coefficient: strong correlation, relatively strong correlation, weak correlation, no correlation, and negative correlation, forming a database of parameter association combinations that can be directly accessed. For strongly correlated and negatively correlated variables, the system automatically performs pairwise pairwise analysis, generating coordinated adjustment schemes in advance to prevent chain reactions caused by single-parameter adjustments, thus achieving overall coordination and closed-loop control of process parameters. For example... Figure 3 As shown, the system dynamically visualizes the correlation between key parameters and solids content (Y) in the form of a heatmap. The color gradient from red to green corresponds to the intervals of negative correlation to strong positive correlation, intuitively reflecting the degree of coupling and changing trend of each parameter with Y over different time periods. By observing this graph in real time, operators can quickly identify the main parameters affecting Y fluctuations and their dynamic changes. When the system detects abrupt changes or abnormal deviations in correlation, it will automatically trigger an early warning, providing a reliable basis for process parameter adjustment and quality prediction. In addition, the system can also output the key characteristic parameters most correlated with Y in real time for subsequent curve fitting, trend prediction, and modeling verification, providing data support and decision-making basis for achieving online precise control and process optimization.

[0054] Standard curve of solids content (Y) versus time: Data Statistics and Scatter Point Calibration: To establish a standard curve of solid content (Y) versus time, the monitoring data from multiple batches of production processes were first statistically calibrated. Due to external factors such as batch differences, noise interference, and random disturbances, data collected at the same process and time point may fluctuate. To ensure the representativeness and stability of the fitted data, the data analysis-based TCM extraction quality management system horizontally aggregated no fewer than 30 batches of data at each time point, obtaining the normal distribution interval of the data at each time point based on the central limit theorem. Subsequently, the data mean (i.e., the normal value) and the 95.4% confidence interval (representing the data fluctuation range) were calculated as the fitting input and boundary constraints for subsequent curve regression. For different data distribution characteristics, the system used cubic spline or smoothed spline methods to smooth the interval boundaries, effectively eliminating noise while maintaining the original dissolution trend, resulting in a continuous and differentiable preliminary curve sequence.

[0055] Standard curve establishment and model validation: Based on smoothed data, a multiple linear regression model was constructed to analyze the linear relationship between multiple independent and dependent variables, and the least squares method was used to solve for the polynomial parameters. Subsequently, the control requirements of key parameters in the pharmacopoeia standards, process specifications, and operation manuals were converted into numerical constraints, and combined with the total solids index to form hard constraint boundaries. On this basis, the area under the standard curve was integrally validated (to verify the consistency between extraction rate and production amount through integration), and the coefficient of determination (R²) and root mean square error (RMSE) of the model were evaluated simultaneously. By iteratively optimizing the polynomial function parameters, it was ensured that the fitting results met the standards in terms of both shape rationality and total closure, ultimately forming a standard curve of solids content versus time.

[0056] Model Storage and Online Application: The generated standard curves are stored in the database as function expressions and parameter sets, and can be dynamically retrieved and visualized based on conditions such as time and batch. Simultaneously, the data analysis-based TCM extraction quality management system synchronously labels the normal value points, ±10% control thresholds, and confidence band ranges on the curve graphs, providing real-time references for online monitoring, predictive warnings, and process optimization. This model can directly serve the dynamic control and auxiliary adjustment decisions of solid content (Y) in the production process, realizing the transformation from post-production detection to real-time prediction and proactive control, providing strong technical support for process stability and product quality consistency.

[0057] S4: Train the prediction model constructed in S3 and initialize the system model library, and define the solid content control threshold.

[0058] To achieve dynamic predictability and early intervention of solid content (Y), a data-driven TCM extraction quality management system constructs a multi-stage prediction model architecture based on the results of previous batch fitting curves and parameter correlation analysis. In the initial stage of the system, with a small sample size, low variable dimensionality, and focusing only on the temporal evolution of Y itself, a differential autoregressive moving average (ARIMA) model is used for short-term prediction modeling. This model can quickly output predicted values ​​and confidence intervals for several future steps, providing a reference for early intervention in the production process. As the system operates and data gradually accumulates, the data-driven TCM extraction quality management system will switch to a Long Short-Term Memory (LSTM) network model to achieve deep learning modeling of multivariate time-series data. This model can simultaneously handle nonlinear relationships, time lags, and multi-parameter interaction effects. Through self-learning, it determines which historical period and which variables are most critical to the future changes of Y, thereby achieving multi-step advance prediction and confidence band output. Meanwhile, the data analysis-based TCM extraction quality management system takes into account the source of raw materials, retention rate of active ingredients, and batch quality differences between different suppliers and batches in real time. It dynamically calculates the mapping relationship between extraction rate and yield and evaluates online what operating conditions the production line should operate under to ensure that the Y value remains stable within the ±10% envelope range.

[0059] S5: Real-time acquisition of production process data, and extrapolation of the changing trend of solid content based on the standard curve established in S3 and the prediction model trained in S4; S6: When real-time monitoring data or prediction results exceed the control threshold, generate multiple process adjustment schemes.

[0060] This data-driven quality management system for traditional Chinese medicine extraction couples information from two aspects: the correlation strength of parameter Y and model prediction error. This jointly pinpoints the real-time impact of each parameter on Y and constructs a multi-objective optimization constraint set. The control rules for key parameters in pharmacopoeias, process requirements, and work manuals are transformed into structured process parameter ranges. The system specifies that adjustments must meet production requirements, serving as a hard constraint for optimization. The experience of frontline operators is quantified as adjustment costs (the larger the adjustment range and the higher the frequency, the higher the cost), serving as a soft constraint. A very small value is added or subtracted from the standard curve envelope to limit the limits of control adjustment schemes during production, ensuring that the adjustment results of the predicted curve and key parameters do not exceed the normal range due to noise and random disturbances. Under the four-dimensional objective constraints of minimum cost, minimum prediction error, Y content range, and physical range of key parameters, the system solves for the optimal solution of local hyperparameters, generating an auxiliary control strategy that can be directly deployed. During production line operation, the data analysis-based TCM extraction quality management system compares the monitoring data, standard curve, and predicted trajectory in real time. Once a deviation is detected, feedback is immediately triggered, driving actuators such as temperature control, pressure, and flow control back to a safe state, and providing regression time estimates and stable state control parameters.

[0061] The data-driven quality management system for traditional Chinese medicine extraction constructs a dual-threshold alarm mechanism based on the solid content standard curve, real-time monitoring values ​​of quality indicator Y, and the fluctuation range of the prediction model. Through comprehensive analysis of historical data distribution characteristics and model prediction deviations, the system automatically sets high and low alarm values, enabling dynamic monitoring and prediction of fluctuations in the production process. When actual monitoring data or prediction results exceed the set threshold range, the system automatically triggers local anomaly trend analysis and generates global anomaly warning information, prompting operators to adjust or intervene in the process. This mechanism can identify potential fluctuation factors affecting the stability of solid content (Y) in advance, shifting from post-event detection to pre-event warning, and improving the predictive control and quality assurance capabilities of the production process.

[0062] S7: Record the execution effect of process adjustments, write the data back to the model library, optimize the prediction model and adjustment strategy, and achieve continuous process optimization.

[0063] The data-driven TCM extraction quality management system provides four steps to assist decision-making: First, it provides early warnings based on rolling forecasts, indicating when and under what conditions parameters are about to exceed limits. Second, it calculates when intervention should terminate and when parameters should stabilize, using the area under the historical and predicted curves (integral) as criteria, ensuring the integral value remains within the allowable fluctuation range of Y throughout the process. Third, it visualizes the short-term trend curves of multiple adjustment schemes, helping operators intuitively grasp the short-term future trends of each parameter for direct comparison. Fourth, it optimizes the scheme by automatically recording the selected parameters and their selection rate after each scheme is adopted, feeding the adjusted parameters and optimization weights back to the decision model to continuously optimize the next round of recommendation ranking. This aims to provide the most easily adjustable and minimally sized emergency fine-tuning path for parameter weights and variable combinations, reducing the intensity of manual intervention.

[0064] After production, the data-driven quality management system for traditional Chinese medicine extraction performs in-depth analysis of historical process data to achieve post-production optimization and anomaly tracking. Utilizing sliding time window technology, it dynamically scans and calculates fixed-length time intervals, extracting the entropy change trends of data within each time period to quantify the degree of system disorder caused by extraction rate fluctuations at different stages, thereby accurately locating the time and scope of anomalies. Based on the established correlation analysis model, the system compares and analyzes the fluctuation characteristics within the abnormal time period with actual operating conditions and production data, tracing the change trajectories of key indicators in each process step. Through anomaly detection models, it identifies key process parameters causing extraction rate fluctuations, and then adjusts equipment operating parameters or process flows accordingly to achieve continuous optimization and improvement of the production process.

[0065] This method identifies key quality control points affecting solid content at each stage by analyzing the entire extraction and concentration process. It simultaneously assesses the online monitoring capabilities of existing equipment (e.g., monitored parameters, missing parameters) and develops a key process parameter acquisition plan to achieve real-time acquisition, transmission, and dynamic control. Based on production and experimental data, a standard curve of solid content (Y) versus production time (online refractometer recommended for solids, near-infrared spectroscopy recommended for spectral data) is established to clarify the theoretical range of Y at different stages. Through real-time monitoring and data correlation analysis of key process parameters (X, such as extraction temperature), a regression model of Y=f(X) is proposed to achieve stable control of solid content (Y), with fluctuations controlled within ±10%. The aim is to build a data-driven quality management system for traditional Chinese medicine extraction that integrates key process parameter monitoring, data storage, anomaly early warning, quality traceability, and analysis optimization. It focuses on the core pain point of excessive reliance on post-production sampling in traditional Chinese medicine extraction and oral liquid production lines, which makes pre-production prediction and in-process intervention difficult, leading to large fluctuations in solid content (Y). To ensure batch-to-batch consistency and stable production, online closed-loop verification of process parameter X and quality indicator Y is implemented, strictly controlling the quantitative threshold of solid content. Simultaneously, dormant data is brought to life, becoming knowledge that guides parameter adjustments. This transforms the traditional passive quality control model of post-production sampling and finished product inspection into a proactive management model of pre-production prediction and in-process intervention. Through multi-stage predictive models, trends in solid content changes are projected in advance, combined with a high and low dual-threshold alarm mechanism, quality risks are identified early, and targeted adjustment plans are generated, significantly reducing the generation of batches of non-conforming products and lowering waste of raw materials and production energy for traditional Chinese medicine.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A data analysis-based method for quality management of traditional Chinese medicine extraction, characterized in that... Includes the following steps: S1: Analyze the entire process of traditional Chinese medicine extraction and concentration, identify key quality control points that affect solid content, and formulate a key process parameter collection plan; S2: Collect multi-source heterogeneous production data, preprocess the collected data, and obtain a high-quality training dataset; S3: Based on the high-quality training dataset described in S2, establish a standard curve of solid content changing with production time, and construct a prediction model after completing the correlation analysis of key parameters. S4: Train the prediction model constructed in S3 and initialize the system model library, and define the solid content control threshold; S5: Real-time acquisition of production process data, and extrapolation of the changing trend of solid content based on the standard curve established in S3 and the prediction model trained in S4; S6: When real-time monitoring data or prediction results exceed the control threshold, multiple process adjustment schemes are generated; S7: Record the execution effect of process adjustments, write the data back to the model library, optimize the prediction model and adjustment strategy, and achieve continuous process optimization.

2. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 1, characterized in that... The multi-source heterogeneous production data mentioned in S2 includes the original process parameter data of the DCS system, the supplementary process parameter data collected by the newly added sensors, and the real-time data of solid content and active ingredients collected by the online detection equipment.

3. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 2, characterized in that the preprocessing in S2 adopts a three-level cleaning method, wherein the three-level cleaning is to sequentially remove outliers, fill in missing values, and replace duplicate values.

4. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 3, characterized in that... The method for establishing the standard curve of solid content change with production time as described in S3 is as follows: S31: At each time point, aggregate no less than 30 batches of historical production data horizontally, and calculate the mean and 95.4% confidence interval of the data at each time point; S32: The interval boundaries are smoothed using cubic spline method or smooth spline method to obtain a preliminary curve sequence; S33: Construct a multiple linear regression model, use the least squares method to solve for the polynomial parameters, and combine the pharmacopoeia standards and process specifications to form hard constraint boundaries; S34: Perform integral verification on the area under the standard curve, evaluate the model's coefficient of determination and root mean square error, and iteratively optimize to obtain the final solids content and time standard curve.

5. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 4, characterized in that... The key parameter correlation analysis described in S3 specifically involves: using Pearson correlation coefficient, cosine similarity, and feature ablation test algorithms to calculate and classify the correlation strength between each key process parameter and solid content; using principal component analysis to reduce the dimensionality of high-dimensional data; and using dynamic time warping algorithm to align time-lag data.

6. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 5, characterized in that... The solid content control threshold is ±10%.

7. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 6, characterized in that... The prediction model adopts a multi-stage architecture: when the sample size is less than a preset threshold, a differential autoregressive moving average model is used for short-term prediction; when the sample size reaches the preset threshold, the model is switched to a long short-term memory network model for multivariate time-series prediction.

8. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 7, characterized in that... Based on the fluctuation range of the solid content standard curve and the prediction model, a dual-threshold alarm mechanism with high and low thresholds is constructed.

9. The method for quality management of traditional Chinese medicine extraction based on data analysis according to claim 8, characterized in that... The process adjustment scheme includes key parameter adjustment values, predicted improvement range of solid content, and confidence level, which are obtained by constructing and solving a multi-objective optimization constraint set that includes the physical range of process parameters, adjustment cost, and solid content control requirements.

10. A data analysis-based quality management system for traditional Chinese medicine extraction, characterized in that... The system operates using the data analysis-based quality management method for traditional Chinese medicine extraction as described in any one of claims 1 to 9, including: The key process parameter centralized control platform is used to realize data acquisition, real-time monitoring, anomaly early warning and historical traceability of various key process parameters in the production process; The solids content modeling and analysis system is used to establish a mechanistic correlation model between solids content and time and process parameters, so as to achieve the prediction, evaluation and optimization control of quality indicators.

Images