Accurate detection system for medicinal liquor components

Abstract

The invention relates to the technical field of medicinal liquor component detection, and discloses an accurate detection system for medicinal liquor components. The system comprises a multi-source spectrum acquisition module, a dynamic component modeling module, a multi-mode difference fusion module, an abnormal component positioning module and a detection strategy optimization module. The multi-source spectrum acquisition module synchronously captures near-infrared absorption spectrum, Raman scattering spectrum and liquid chromatogram-mass spectrum combined data, and generates a standardized spectrum data matrix through adaptive filtering; the dynamic component modeling module constructs a feature extraction network based on a historical spectrum library, and outputs a theoretical component vector containing a target active component concentration interval; the multi-mode difference fusion module generates a multi-dimensional difference tensor through multi-mode difference analysis; the abnormal component positioning module generates an abnormal component distribution probability thermodynamic diagram in combination with the medicinal material association topology network; and the detection strategy optimization module dynamically configures detection parameters according to the thermodynamic diagram. According to the system, accurate detection and abnormal positioning of medicinal liquor components are realized, and the comprehensiveness and intelligent level of detection are improved.

Description

Technical Field

[0001] This invention relates to the field of medicinal wine component detection technology, specifically a precise detection system for medicinal wine components. Background Technology

[0002] As an important dosage form of traditional Chinese medicine, medicinal wines present stringent requirements for detection technologies due to the complexity and diversity of their components. Currently, traditional detection methods mostly rely on single spectroscopic techniques or chromatographic analysis, which have significant limitations. For example, while near-infrared spectroscopy can quickly acquire sample information, it is easily affected by ambient light fluctuations, leading to a decrease in the signal-to-noise ratio of the spectral signal and making it difficult to accurately capture the characteristic peaks of trace active ingredients. In the process of Raman scattering spectroscopy, characteristic peaks often overlap due to sample matrix interference, making it impossible to effectively distinguish structurally similar isomers.

[0003] While liquid chromatography-mass spectrometry (LC-MS) offers high component separation capabilities, it consumes large amounts of organic solvents during detection and has a long single-detection cycle, making it difficult to meet the needs of rapid screening of large-scale samples. Furthermore, existing detection systems lack a deep fusion mechanism for multi-source data; data acquired by different detection methods are often processed independently, leading to biases in component analysis results.

[0004] Traditional detection methods rely heavily on manual experience to locate abnormal components, lacking systematic analysis based on the medicinal material's origin and the interactions between its components. This makes it difficult to accurately identify abnormal components caused by adulteration or deterioration. These shortcomings not only affect the accuracy of medicinal wine quality evaluation but also hinder the innovative development of traditional dosage form quality control technologies in the modernization of traditional Chinese medicine. Summary of the Invention

[0005] The purpose of this invention is to provide a precise detection system for medicinal wine components to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a precise detection system for medicinal wine components, the system comprising:

[0007] Multi-source spectral acquisition module: Simultaneously captures the near-infrared absorption spectrum sequence, Raman scattering spectrum temporal characteristics, and liquid chromatography-mass spectrometry data stream of medicinal wine samples, eliminates ambient light interference through adaptive filtering, and generates a standardized spectral data matrix;

[0008] Dynamic component modeling module: Based on the historical medicinal material spectral library, a dynamic feature extraction network is constructed. The standardized spectral data matrix is ​​input into the feature extraction network, and the output is a theoretical component vector including the concentration range of the target active ingredient.

[0009] Multimodal difference fusion module: Performs multimodal difference analysis on the theoretical component vector and the reference component vector of the standard medicinal material database. The multimodal difference analysis includes spectral absorption peak shift, characteristic peak intensity ratio deviation and chromatographic retention time distribution difference, and generates component-level multidimensional difference tensors.

[0010] Abnormal component localization module: Input the multidimensional difference tensor into the medicinal material association topology network, combine the medicinal material source similarity matrix and component interaction parameters to generate an abnormal component distribution probability heat map and identify abnormal component aggregation areas;

[0011] The detection strategy optimization module dynamically configures detection parameters based on the gradient distribution of the probability heatmap of the abnormal components, including: enabling high-resolution microfluidic chip resampling for high-probability abnormal areas; and applying cross-modal spectral perturbation testing to the associated medicinal components.

[0012] Preferably, the multi-source spectral acquisition module specifically includes:

[0013] Heterogeneous sensor array: Deploys a collaborative acquisition unit of UV-Vis spectrophotometer, Fourier transform infrared spectrometer and mass spectrometer detector to simultaneously acquire spectral absorption curves, Raman shift characteristic peak intensity distribution and molecular weight distribution chromatogram;

[0014] Time-series feature calibration unit: Performs time-domain alignment processing on the real-time acquired spectral data stream, extracts the peak drift rate and half-width variation parameters of the absorption spectrum, the signal-to-noise ratio and phase angle distribution of the shift characteristic peaks of the Raman spectrum, and the retention time gradient and peak area entropy value of the chromatographic peaks.

[0015] Noise suppression submodule: It uses wavelet packet transform to decompose baseline drift noise caused by ambient temperature and humidity, and eliminates high-frequency random interference through adaptive threshold filtering.

[0016] Preferably, the dynamic component modeling module further includes executing:

[0017] Based on the feature weighted fusion of the compatibility rules of medicinal materials, the correlation features of near-infrared spectroscopy, Raman fingerprinting and mass spectrometry fragments are integrated;

[0018] Multispectral cross-modal correlation feature vectors are extracted using a convolutional attention mechanism;

[0019] The output includes a theoretical component vector containing the source identifier of the medicinal material, and the theoretical component vector is dynamically updated with each batch of medicinal materials.

[0020] Preferably, the multi-modal difference fusion module specifically includes:

[0021] Spectral offset calculation: The absorption spectra of the theoretical component vector and the reference component vector are compared by sliding with a preset wavelength window. The asynchronously sampled spectral sequence is aligned by a dynamic time warping algorithm. The cumulative offset of characteristic peaks in each window is calculated to generate a spectral offset vector.

[0022] Characteristic peak intensity assessment: Perform band energy decomposition on the Raman characteristic peaks of the theoretical vector and the reference vector, calculate the energy density ratio of the main peak to the secondary peak, extract the intensity deviation index of each displacement interval, and construct the intensity deviation vector;

[0023] Chromatographic distribution difference detection: Based on the structural similarity algorithm, the time distribution of theoretical values ​​and reference chromatographic peaks is matched, the phase error of retention time abrupt change points is calculated, the statistical divergence of peak area distribution is quantified, and a chromatographic difference vector is generated;

[0024] Multidimensional difference tensor generation: The spectral offset vector, intensity deviation vector, and chromatographic difference vector are spliced ​​into three-dimensional tensors, and normalized by feature significance weighting to output a third-order tensor with the dimension of [component number × detection batch × difference type].

[0025] Preferably, the abnormal component localization module specifically includes:

[0026] Topological modeling of medicinal materials: Construct a component association topology map based on the source information of medicinal materials, mark the chemical bond connection relationship and molecular interaction parameters between each component, and superimpose the solvent effect constraint of the medicinal wine preparation process on the topology map to generate a chemical topology model including molecular similarity matrix and component affinity matrix;

[0027] Anomaly propagation simulation: The multidimensional difference tensor is mapped to the corresponding node of the chemical topology model, and anomaly component diffusion inference is performed based on graph neural network. The diffusion inference includes: calculating the migration attenuation factor of the anomaly component based on molecular interaction parameters; capturing cross-component anomaly correlation features through multi-head attention mechanism; and simulating the diffusion path of the anomaly component in the topology network using Monte Carlo method.

[0028] Heatmap generation: Statistically count the frequency of abnormal occurrences of each chemical component in simulated diffusion, calculate the residence probability value of abnormal components by combining molecular affinity parameters, and generate an anomaly probability heatmap covering the entire component set;

[0029] Physical component localization: Perform spatial clustering analysis on the anomaly probability heatmap to identify the set of chemical components whose anomaly probability exceeds a preset threshold.

[0030] Preferably, the detection strategy optimization module specifically includes:

[0031] High-resolution resampling trigger unit: When the probability value of a specific region in the abnormal probability heatmap exceeds the first threshold, an instruction is sent to the microfluidic chip corresponding to that region to execute: increase the spectral sampling resolution to 5 times the original value; simultaneously enable the mass spectrometry multi-reaction monitoring mode to capture the characteristic ion fragment intensity ratio; deploy a transient peak shape recorder at the detection end to store abnormal chromatographic peak fragments;

[0032] Spectral perturbation test unit: Apply multi-band spectral excitation to the associated chemical components with the largest gradient changes in the anomaly probability heatmap.

[0033] Preferably, the abnormal component localization module also outputs a set of suspicious component identifiers and an abnormal propagation main path;

[0034] The set of suspicious component identifiers binds chemical components to medicinal material sources based on the clustering results of the anomaly probability heatmap;

[0035] The main path of anomalous propagation is generated by recording the most frequent component migration sequences in Monte Carlo simulations.

[0036] Preferably, the system further includes:

[0037] Component risk rating module: Calculates the anomaly confidence level of each chemical component based on the distribution gradient of the anomaly probability heatmap;

[0038] Based on the historical adverse reaction database, anomaly confidence levels and safety thresholds are matched to generate component risk level labels;

[0039] The test report generation module integrates the component risk level labels, the probability heatmap of abnormal component distribution, and the multidimensional difference tensor to output a structured test report.

[0040] Preferably, the detection strategy optimization module further includes:

[0041] The flow rate parameters of the microfluidic chip are dynamically adjusted according to the risk level label of the ingredients.

[0042] When a high-risk component is detected, the traceability procedure for medicinal materials is automatically triggered.

[0043] Preferably, the component risk rating module further includes the following:

[0044] When a chemical component with an abnormal confidence level exceeding the safety threshold is detected, a risk component interception signal is generated; based on the risk component interception signal, an automatic fractionation device is triggered to perform a physical isolation operation, and the abnormal component batch mark in the medicinal material traceability database is updated synchronously.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] The multi-source spectral acquisition module enables the simultaneous acquisition of near-infrared absorption spectroscopy, Raman scattering spectroscopy, and liquid chromatography-mass spectrometry data. Combined with adaptive filtering technology, ambient light interference is effectively eliminated, resulting in a standardized spectral data matrix that more accurately reflects the component characteristics of the sample, providing a reliable data foundation for subsequent analysis. The dynamic component modeling module, based on a dynamic feature extraction network constructed from a historical medicinal herb spectral database, can transform the standardized data matrix into a theoretical component vector containing the concentration range of the target active ingredient. This achieves accurate analysis of complex component systems, overcoming the limitations of traditional single-model component prediction.

[0047] The multimodal difference fusion module performs multimodal difference analysis on the theoretical component vector and the reference vector in the standard database, covering dimensions such as spectral absorption peak shift, characteristic peak intensity ratio deviation, and chromatographic retention time distribution differences. The generated multidimensional difference tensor comprehensively presents the subtle differences between the sample and the standard, improving the comprehensiveness and detail of component comparison. The abnormal component localization module inputs the multidimensional difference tensor into the medicinal material association topology network, and combines it with the medicinal material source similarity and component interaction parameters to generate an abnormal component distribution probability heatmap. This can intuitively identify the aggregation areas of abnormal components, eliminating the subjectivity of traditional manual judgment and making abnormal component identification more scientific and accurate.

[0048] The detection strategy optimization module dynamically adjusts detection parameters based on the gradient distribution of the heatmap, uses a high-resolution microfluidic chip for resampling in high-probability anomaly areas, and applies cross-modal spectral perturbation testing to related medicinal components. This improves detection efficiency while ensuring accuracy and avoiding unnecessary resource consumption. The entire system achieves intelligent processing throughout the entire process, from data acquisition and analysis to anomaly localization and strategy optimization. It adapts to the actual needs of complex medicinal wine components and high detection requirements, providing a new technical approach for the quality control of medicinal wine. Attached Figure Description

[0049] Figure 1 This is a schematic diagram illustrating the working principle of the precise detection system for medicinal wine components described in this invention.

[0050] Figure 2 Flowchart for the dynamic component modeling module;

[0051] Figure 3 A flowchart for the multi-modal difference fusion module;

[0052] Figure 4 Flowchart for the abnormal component localization module;

[0053] Figure 5 A flowchart for generating component risk ratings and reports. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] Please see Figure 1 This invention provides a precise detection system for medicinal wine components, the system comprising:

[0056] By employing multi-source spectral collaborative analysis, dynamic component modeling, and intelligent anomaly localization, the system achieves accurate detection and risk warning of medicinal wine components. The system uses a multi-source spectral acquisition module to simultaneously capture near-infrared absorption spectra, Raman scattering spectra, and liquid chromatography-mass spectrometry data from medicinal wine samples. A dynamic component modeling module generates theoretical component vectors, which are then compared with a standard medicinal material database through a multi-mode difference fusion module to output a multi-dimensional difference tensor. An anomaly localization module generates an anomaly probability heatmap based on the medicinal material association topology network. A detection strategy optimization module dynamically adjusts detection parameters according to the heatmap, achieving high-precision component analysis.

[0057] Example 1: See Figure 2 The multi-source spectral acquisition module employs a heterogeneous sensor array to capture the multi-dimensional optical features of the medicinal wine samples. The UV-Vis spectrophotometer's optical path system uses a combination of deuterium and tungsten halogen lamps as light sources, equipped with a grating monochromator and a silicon photodiode array detector. During scanning, a stepper motor drives the grating rotation to achieve full-band coverage. The Fourier transform infrared spectrometer is equipped with a diamond ATR accessory, and the interferometer's moving mirror scanning speed is automatically adjusted according to the sample's absorption characteristics. A liquid nitrogen-cooled MCT narrowband detector is used. The mass spectrometer detector employs an orthogonal accelerated time-of-flight mass analyzer, and the ion source temperature is maintained within a suitable range to avoid sample thermal decomposition.

[0058] Synchronous acquisition of the sensor array is achieved through hardware trigger signals. The main control unit sends synchronization pulses to each instrument, and the data acquisition card records the raw signals with a unified timestamp. When processing asynchronous data streams, the timing feature calibration unit first establishes a time offset model between instruments, calculates the clock deviation of each device using a cross-correlation algorithm, and then achieves data resampling alignment through cubic spline interpolation. Peak drift detection in near-infrared spectroscopy uses a sliding window extreme value localization algorithm, and the half-peak width variation parameter is characterized by the slope of the linear fitting of the high points of the left and right halves of the peak. The signal-to-noise ratio evaluation of Raman spectroscopy is performed in the baseline region near the characteristic peak. The noise standard deviation is calculated in the interval without characteristic peaks, and the instantaneous phase information of the phase angle distribution is extracted through Hilbert transform.

[0059] The noise suppression submodule employs a multi-stage filtering architecture to eliminate environmental interference. Wavelet packet transform decomposition selects a mother wavelet function suitable for the spectral characteristics, and the number of decomposition levels is dynamically determined based on the noise spectrum distribution. Baseline drift correction is achieved through approximate coefficient reconstruction, preserving low-frequency trend terms while suppressing high-frequency noise. Adaptive threshold filtering uses a bivariate shrinkage algorithm, where the local threshold window size is negatively correlated with signal fluctuation intensity, and morphological opening operations are used to handle sudden impulse noise. The temperature and humidity compensation unit incorporates an environmental sensor and establishes a temperature drift model to correct detector response deviations in real time.

[0060] The convolutional attention network of the dynamic component modeling module is designed with a multi-branch structure. The near-infrared spectroscopy branch uses a one-dimensional convolutional kernel that slides along the wavelength dimension, followed by batch normalization and the LeakyReLU activation function. The Raman spectroscopy branch focuses on fingerprint region features, using dilated convolutions to expand the receptive field and capture a wide range of shift patterns. The mass spectrometry branch employs a residual connection structure to avoid gradient vanishing in deep networks, and introduces a nonlocal attention mechanism to capture long-range correlations for fragment ion peak processing. In the feature fusion stage, a cross-modal attention gating system is designed, balancing the contribution of each spectrum through a learnable weight matrix, and the attention score is calculated using a scaled dot product.

[0061] The network training employs a multi-task learning strategy. The main task outputs a theoretical component vector, while auxiliary tasks include medicinal material source classification and efficacy component prediction. The loss function is designed as a combination of weighted mean squared error and cross-entropy. The optimizer uses an adaptive moment estimation algorithm with a gradient pruning threshold. An online update mechanism deploys an incremental learning framework. When new batches of data are input, an elastic weight consolidation algorithm protects important parameters, and the degree of historical knowledge retention is controlled by the diagonal elements of the Fisher information matrix. The model interpretability module integrates gradient-weighted class activation mapping technology to visualize the contribution distribution of each spectral region to the final decision.

[0062] The medicinal herb spectral library is constructed using a hierarchical storage structure. Raw spectral data is preprocessed and stored in a time-series database, while feature vectors are stored in a graph database to establish a medicinal herb-component association network. A data augmentation module applies appropriate spectral perturbations to rare medicinal herb samples, including Gaussian noise injection, wavelength shift simulation, and intensity ratio adjustment. A version control system records each database update and supports backtracking to historical states by timestamp. The query interface provides multi-dimensional search functionality, supporting queries based on combinations of conditions such as chemical composition, medicinal herb origin, and collection time.

[0063] The hardware system adopts a modular design, with the spectral acquisition unit mounted on a shockproof optical platform. Optical paths are connected between instruments via fiber optic coupling. The temperature control system maintains the sample chamber temperature within a standard range and humidity within a suitable range to prevent sample moisture absorption. The automatic sample introduction device integrates a precision syringe pump and a six-way valve; gradient elution is used during sample switching to avoid cross-contamination. The main control computer is equipped with a real-time operating system, with data acquisition threads running on an independent kernel to ensure timing accuracy. Network communication uses a time-sensitive network protocol to synchronize the clocks of each subsystem.

[0064] The quality control module implements full-process monitoring, including sample injection volume calibration, light source intensity stability testing, and mass spectrometer mass axis calibration. Daily system performance verification tests are performed, including blank solvent analysis, standard repeatability testing, and linearity range verification. The anomaly diagnosis system monitors the status parameters of each instrument in real time and provides tiered early warnings for faults such as voltage fluctuations and flow anomalies. Data integrity checks employ a checksum mechanism, and forward error correction coding is used during transmission to ensure data reliability.

[0065] The user interface design follows the analytical workflow logic, with the main view dynamically displaying a 3D spectral data cube, supporting multi-angle rotation and profile extraction. The results report interface adopts an interactive visualization design, with abnormal component annotations and medicinal material source information displayed in conjunction. The system settings interface provides advanced adjustment options for parameters such as spectral resolution and integration time, and the access control system differentiates between three levels of access control: operators, administrators, and maintenance engineers. The audit trail function records all critical operations, including parameter modifications, data deletion, and system calibration events.

[0066] The maintenance subsystem implements predictive maintenance, monitoring the wear and tear of mechanical components via vibration sensors, and triggering optical component cleaning reminders based on a transmittance decay model. The consumables management system tracks metrics such as column usage count and cumulative light source operating time, generating replacement plans in advance. Remote diagnostics supports secure tunnel connections, allowing manufacturer engineers to view equipment logs and operating parameters. Firmware upgrades employ a dual-partition design, ensuring a rollback to a stable version in case of update failure.

[0067] Example 2: See Figure 3 The multi-mode difference fusion module performs in-depth comparison analysis between theoretical component vectors and standard medicinal material databases. The spectral offset calculation unit uses a dynamic time warping algorithm to process asynchronously sampled spectral sequences, with curved path constraints set in the algorithm implementation to maintain the rationality of characteristic peak correspondences. The wavelength window sliding process employs an overlapping design, retaining partial overlap between adjacent windows to avoid missing boundary features. During offset quantization, an absorption intensity weighting coefficient within the window is introduced, with the characteristic peak offset in strong absorption regions contributing more to the final result. The cumulative offset calculation integrates local differences from multiple windows, and normalization eliminates the dimensional influence of inherent absorption intensities between different components.

[0068] The characteristic peak intensity assessment unit performs frequency band decomposition on the Raman spectrum, with the frequency band division referencing the typical distribution pattern of the characteristic fingerprint region of the medicinal materials. Energy density calculation employs a numerical integration method to process discrete spectral data, and the determination of primary and secondary peaks is based on relative intensity ranking and prior knowledge from the medicinal material database. The intensity deviation index calculation considers the overall distribution morphology of the characteristic peaks, comparing not only the intensity differences of individual peaks but also assessing the synergistic variation trend of multiple characteristic peaks within a specific frequency band. An adjustment factor for the compatibility relationship of the medicinal materials is introduced during the deviation vector construction process, appropriately reducing the deviation weight for component groups with known synergistic effects.

[0069] Chromatographic distribution difference detection employs a multi-scale similarity matching strategy. The structural similarity algorithm considers both the overall peak profile and local details, establishing a non-rigid correspondence along the retention time dimension. Phase error detection identifies peak deformation characteristics through instantaneous frequency analysis, with special handling applied to asymmetric peaks that have steep leading edges and gentle trailing edges. Statistical divergence calculation integrates the first and second moments of peak area distribution, using probability distribution distance to measure the deviation between theoretical and reference values. When generating the difference vector, a time-weighted coefficient is applied to retention time abrupt changes, with more recently occurring abrupt changes exhibiting higher sensitivity.

[0070] The multidimensional difference tensor generation process employs a feature significance weighting mechanism. Significance weights are dynamically adjusted based on the statistical characteristics of historical detection data, assigning higher weights to component-detection type combinations that exhibit recurring differences. Normalization utilizes a piecewise linear transformation method to unify the scale of difference indices across different dimensions. Tensor concatenation preserves the original physical meaning of each difference dimension and arranges them in memory in primary order of component number to optimize subsequent access efficiency. The metadata record of the third-order tensor includes component identifiers, detection batch numbers, and difference type codes, supporting fast retrieval and slicing operations.

[0071] The differential analysis algorithm employs multiple optimization techniques. The spectral sequence alignment process utilizes a sparse matrix implementation based on dynamic programming, significantly reducing memory consumption. Raman band energy calculation employs Fast Fourier Transform to accelerate convolution operations, and the band division template is efficiently applied using bitmasking technology. Chromatographic peak matching introduces a multi-resolution analysis strategy, first quickly locating approximate corresponding regions at low resolution, and then performing fine-grained comparisons at high resolution. Tensor operations leverage the single-instruction multiple-data instruction set of modern processors to process differential calculations of multiple components in parallel.

[0072] Quality control employs multi-level verification. Input data validation includes spectral integrity checks, component vector validity verification, and reference database version compatibility testing. Intermediate results at each stage of the calculation process are monitored and recorded, and an anomaly detection mechanism triggers a review process for discrepancies exceeding reasonable limits. Output tensors undergo consistency verification to ensure correct data index correspondence across dimensions. Performance benchmarking involves regularly running standard sample datasets to monitor changes in the stability of the difference analysis algorithm.

[0073] In terms of system integration, efficient data flow is achieved. The spectral data interface supports parsing of raw data formats from various instrument manufacturers, and component vector transmission uses compact binary encoding to reduce network load. Reference database access implements a local caching mechanism, ensuring that vector data of frequently queried components resides in memory. The difference tensor storage design employs a columnar data structure to optimize the batch processing efficiency of large-scale component sets. Computational task scheduling uses a work-stealing algorithm to balance the load on multi-core processors, and critical path tasks are given execution priorities.

[0074] Visualization features enhance the interpretability of the analysis process. The spectral difference view simultaneously displays the overlay comparison of theoretical and reference spectra, with key offset regions highlighted in color. Raman intensity deviation uses a radar chart to show the relative differences across frequency bands, supporting interactive adjustment of the frequency band range. Chromatographic distribution differences present a 3D surface comparison of retention time and intensity, with the ability to rotate to observe peak details. The difference tensor summary report automatically generates a sorted list of key anomalous components, accompanied by a histogram of difference type distribution.

[0075] Anomaly detection sensitivity adjustment offers multi-level configuration. The global sensitivity parameter controls the overall difference threshold, influencing the final anomaly determination criteria. Component-level adjustment factors allow for personalized sensitivity settings for specific chemical components, suitable for the analysis of known volatile components. Difference type weights can be dynamically adjusted, emphasizing chromatographic differences during method development and spectral characteristics during stable production. The batch-to-batch comparison mode supports deviation analysis of current detection data from historical averages.

[0076] Maintenance and upgrade features ensure continuous system optimization. The differential calculation algorithm supports hot-swap updates, allowing new version logic to be loaded without restarting the service. The reference database maintenance tool provides batch import and incremental update functions, and the version rollback mechanism retains historical data snapshots. The performance analysis module records the execution time of each step, helping to locate computational bottlenecks. The diagnostic mode outputs detailed calculation process logs for algorithm logic verification and anomaly troubleshooting.

[0077] Security measures protect critical data assets. Spectral data access is encrypted for storage and transmission, and component vector transmission uses digital signatures to prevent tampering. Modifications to the reference database require dual authentication, and audit logs record all data changes. Difference calculation tasks run in a sandbox environment, restricting system resource access permissions. Difference results for sensitive components are subject to tiered access control, and core medicinal herb formula data is stored using segmented encrypted storage.

[0078] The human-computer interaction design optimizes the user experience. The batch processing interface supports drag-and-drop file uploads and templated parameter settings. The real-time monitoring dashboard visually displays the progress of difference analysis and resource usage. The results export function provides conversion to multiple industry-standard data formats and supports integration with third-party analysis software. The quick operation menu allows users to save frequently used analysis workflows, and user-defined workflows can be saved as templates for reuse.

[0079] The algorithm update mechanism maintains the sophistication of the analytical methods. The online learning function progressively optimizes the difference calculation parameters to adapt to natural batch variations in medicinal materials. The experimental mode allows for the deployment of new algorithms for shadow runs, with results compared and validated against existing methods. A third-party algorithm integration framework supports the insertion of difference calculation methods that conform to compatible specifications. The community contribution module encourages users to share validated analytical strategy templates.

[0080] The differential visualization and in-depth analysis tool provides multi-faceted insights. 3D scatter plots display the distribution and clustering of components in the differential space, with interactive filtering to focus on areas of interest. Time series analysis presents the epochal trends of specific component differential indicators, supporting sliding window statistics. Overlaying combined differential maps displays key differential features from multiple testing batches, aiding in the identification of systematic shifts. Relationship network diagrams reveal the transmission relationships of component differences, visually presenting the path of anomaly propagation.

[0081] Continuous improvement mechanisms ensure system evolution. User feedback channels collect operational pain points and functional requirements, which are then incorporated into iterative development plans. Pattern analysis tools are used to identify frequently used functions and potential optimization points to guide interface improvements. Benchmark suites regularly assess system performance changes to prevent updates from introducing performance degradation. Technical debt tracking manages and records code modules requiring refactoring, and plans a systematic optimization roadmap.

[0082] A difference analysis knowledge base accumulates domain expertise. A typical difference pattern library categorizes and stores verified anomaly features, accelerating the identification of new detection problems. A solution knowledge graph links anomaly patterns with handling experience, assisting analysts in decision-making. A medicinal herb variation database records component differences across different origins and harvesting seasons, improving reference benchmarks. An automatic summarization function extracts valuable quality control insights from numerous analysis reports.

[0083] Example 3: See Figure 4The abnormal component localization module utilizes a medicinal herb association topology network to achieve deep analysis of multidimensional difference tensors and anomaly propagation modeling. The medicinal herb topology modeling stage first constructs a graph structure representation of chemical components, where nodes represent active ingredients or characteristic markers in the medicinal wine, and edges represent interactions between components. Node attributes include physicochemical parameters such as molecular weight, polarity, and characteristic spectral peak positions, while edge attributes quantify the interaction strength using the following formula:

[0084]

[0085] In the formula: This represents the interaction weight between components i and j. The structural similarity coefficient (values ​​0-1) is used. This represents the number of hydrogen bond interaction sites. The area of ​​the van der Waals interaction (in Ų). This represents the absolute value of the difference in hydrophobic parameters. During the construction of the topology network, the solvent effect constraint is implemented by introducing a virtual environment node. The edge weights of this node and each component reflect their solubility characteristics in the ethanol solution.

[0086] In anomaly propagation simulations, when mapping multidimensional difference tensors to nodes in a topological network, eigenvalue decomposition is used to convert the third-order tensor into initial anomaly scores for each node. The propagation model of the graph neural network contains six layers of gated graph attention units, each layer fusing the anomaly states and interaction parameters of its neighborhood components when computing node representations. Anomaly migration attenuation factor. By edge weight A composite function of node degree is generated to control the propagation rate of anomalous signals in the topological network. The multi-head attention mechanism sets up four independent attention heads, each focusing on different modes of action such as chemical bond connections, spatial conformational similarity, pharmacophore matching, and metabolic pathway association.

[0087] In the Monte Carlo simulation of anomaly propagation, a Markov chain Monte Carlo method is used to generate 1000 independent propagation paths. The transition probability of each path is determined by the product of the anomaly score of the current node and the edge weight, and the path length follows a Poisson distribution. Three types of key events are recorded during the simulation: the number of anomalous activations of component nodes. Abnormal crossing frequency of boundaries and the duration of the abnormal state. In the heatmap generation stage, discrete events are converted into continuous probability distributions through kernel density estimation, and the bandwidth selection adopts the Silverman adaptive rule.

[0088] Physical component localization employs a density-based spatial clustering algorithm to process probabilistic heatmaps. Chemical spatial constraints are introduced during clustering by embedding molecular descriptors into Euclidean distance calculations to ensure that spatially adjacent components have similar physicochemical properties. Outlier region boundaries are extracted using a level set method, with a probability threshold applied. The settings are dynamically adjusted based on the medicinal material quality control standards, with the default setting being the 95th percentile of the probability distribution. The location results are output as a set of abnormal components. Each element contains an ingredient identifier, an anomaly probability value, and information about the source of the associated medicinal materials.

[0089] The high-resolution resampling trigger unit of the detection strategy optimization module is designed as a state machine model. When the heatmap area... satisfy hour( (With a preset threshold), the system switches to high-precision detection mode. The microfluidic chip's flow rate control employs a PID algorithm, increasing the sampling resolution to five times that of the standard mode, while simultaneously activating secondary fragment ion scanning in the mass spectrometry multi-reaction monitoring mode. A transient peak recorder captures abnormal chromatographic peak waveforms at a 100MHz sampling rate, with a storage depth sufficient for recording at least three complete peak shapes. The spectral perturbation test unit generates multi-band excitation signals, with frequency selection adhering to the following principles: covering the characteristic absorption bands of abnormal components, avoiding regions with strong solvent interference, and including at least one reference standard peak.

[0090] The linked query function with the medicinal material source database enhances the interpretability of anomaly localization. When a high-probability anomalous component is identified, the system automatically retrieves the standard content range of that component in the Chinese Pharmacopoeia, its typical distribution in medicinal materials from different origins, and the characteristic differences of common adulterants. This related information is presented in the form of a knowledge graph, helping analysts distinguish whether the anomaly is caused by process deviations, raw material quality issues, or adulteration.

[0091] The real-time monitoring interface provides dynamic visualization of the topology network. Users can interactively explore the interconnected paths between component nodes and view detailed parameters of any edge. The anomaly propagation process is simulated using a particle system; particle color represents anomaly intensity, and particle trajectory reflects the main diffusion direction of the Monte Carlo simulation. The heatmap overlay display function supports comparing current detection results with historical benchmarks, with areas of difference highlighted as contour lines.

[0092] In terms of algorithm optimization, the training of the graph neural network employs negative sampling to address the imbalance problem of medicinal component categories. For toxic components or trace active substances that occur less frequently, a higher misclassification penalty weight is assigned to the loss function. Network parameter updates utilize stochastic gradient descent with momentum, and the learning rate is dynamically adjusted based on the accuracy on the validation set. To prevent overfitting, the dropout layer randomly masks 20% of neuron connections, and L2 regularization constraints are applied to the hidden layers.

[0093] The anomaly report generation module outputs structured detection results, containing three levels of information: a summary layer listing key anomalies and their risk levels; a detailed layer displaying propagation paths and interaction evidence within the topological network; and a raw data layer providing downloadable probability heatmap matrices and differential tensor slices. The report signature employs an elliptic curve cryptography-based digital signature algorithm to ensure the results are tamper-proof and their origin verifiable.

[0094] Example 4: See Figure 5 The abnormal component localization module outputs a set of suspicious component identifiers, which are then linked to the medicinal material source for precise traceability. Taking the testing of a batch of ginseng medicinal wine as an example, the system identified three components with a high probability of abnormality: ginsenoside Rb1, maltol, and isoimperatorin. By querying the medicinal material source database, the system establishes a correlation between chemical components and raw medicinal materials. Ginsenoside Rb1 is linked to the root of ginseng (Panax ginseng), a plant in the Araliaceae family; maltol is associated with barley germination products; and isoimperatorin corresponds to Angelica dahurica (Angelica dahurica), a plant in the Apiaceae family. The linking process employs a two-way verification mechanism, both by reversely searching for possible source medicinal materials through chemical components and by forward matching feature markers based on the known component list of the medicinal materials.

[0095] The generation of the suspicious component identifier set involves multi-dimensional verification steps. Each anomalous component must meet the following conditions to be included in the final report: clearly recorded in the medicinal material source database, with an anomaly probability value exceeding a set threshold, and showing a traceable trend in historical testing data. The system output format uses a structured table containing complete key information fields:

[0096] Table 1: Correlation chart between chemical components and raw medicinal materials of a certain batch of ginseng medicinal wine.

[0097] Ingredient name Chemical formula medicinal material source abnormal probability Safety threshold Risk level Historical fluctuation range Ginsenoside Rb1 C54H92O23 Araliaceae ginseng 0.93 1.2 mg / mL High risk 0.8-1.1 mg / mL maltol C6H6O3 Barley malt 0.87 0.05 mg / mL Medium risk 0.02-0.04 mg / mL Euphorbia pekinensis C16H14O4 Angelica dahurica (Apiaceae family) 0.78 0.3 mg / mL Low risk 0.1-0.25 mg / mL

[0098] Anomaly propagation path analysis reveals the potential transmission mechanism of component abnormalities. Referring to Table 1, in the above ginseng medicinal wine case, Monte Carlo simulation shows that the most frequent propagation path is: maltol → isoimperatorin → ginsenoside Rb1. Path analysis combined with process knowledge base interpretation suggests possible quality problem points: improper temperature control during barley malt processing leads to an abnormally high maltol content, which in turn affects the solubility characteristics of angelica extract, ultimately altering the dissolution efficiency of ginsenosides. The main path visualization uses a directed graph, where node size reflects the probability of anomalies, edge width represents propagation intensity, and critical paths are highlighted in red.

[0099] The ingredient risk rating module implements a tiered control strategy. Risk levels are classified based on three core dimensions: calculated anomaly confidence level, deviation from the pharmacopoeia-stipulated safety threshold, and the ingredient's critical weight in product efficacy. High-risk ingredients trigger an immediate halt to production control signals; medium-risk ingredients require batch isolation for further investigation; and low-risk ingredients record deviations but are allowed to continue production. An expert review mechanism is incorporated into the rating process. When there is a significant difference between the system's automatic rating and historical manual assessments, the results are automatically sent to quality control engineers for final determination.

[0100] The test report generation module integrates multi-dimensional data output for comprehensive evaluation. The structured report comprises six main parts: a summary page presenting key anomalous components and overall risk conclusions; a heatmap display page supporting interactive zooming to view local details; a difference analysis page comparing the spectral and chromatographic characteristics of the current batch with standard batches; a process impact page analyzing potential problem areas; a corrective action page listing recommended treatment plans; and an appendix containing complete descriptions of test parameters and methods. Report version control uses a naming convention of "year-batch-version number" to ensure uniqueness during traceability.

[0101] The medicinal herb sourcing traceability function enables anomaly root cause analysis. For the high-risk component ginsenoside Rb1, the system automatically initiates a full-chain traceability query: it traces back from the finished product batch to the extract input records, pinpointing the specific ginseng raw material batch; then it queries the warehousing inspection data of that batch of medicinal materials, finding that its saponin content is below the standard lower limit; further tracing back to the planting stage, it links to records of abnormally increased rainfall during the growing season. The traceability results are presented in a timeline format, with key nodes accompanied by scanned copies of the original test reports and the signatures of the responsible persons.

[0102] The risk interception system implements a tiered response mechanism. For the high-risk component ginsenoside Rb1 listed in the table, the following control actions are automatically executed: a shutdown command is sent to the bottling line PLC to isolate the medicinal wine product in the current bottling section; the solenoid valve control of the fractionation unit is activated to direct the relevant fraction to the isolation storage tank; and an emergency event notification is sent to the quality management department, along with a complete anomaly analysis report. The medium-risk component maltol triggers an early warning signal, displaying an anomaly message on the production dashboard and requiring the process engineer to confirm the handling plan. The low-risk component isoimperatorin only records deviations, which are summarized and explained in the monthly quality analysis report.

[0103] The system enhances the clinical relevance of risk assessment through correlation queries in the adverse reaction database. It automatically matches cases of ginsenoside Rb1 exceeding the standard with historical adverse reaction reports, discovering two previous records of gastrointestinal discomfort caused by similar fluctuations in ginsenoside Rb1 levels. This correlation of clinical evidence significantly improves the accuracy of risk rating and provides a reference direction for corrective measures. Adverse reaction data is stored anonymously and includes structured fields such as symptom description, frequency of occurrence, and dose-related factors, supporting multi-condition combined queries.

[0104] A dynamic threshold adjustment mechanism addresses natural variations in medicinal materials. Ginsenoside content typically exhibits seasonal fluctuations. The system establishes a monthly adjustment coefficient model by analyzing test data from the same period over three consecutive years. For ginseng raw materials harvested in September, the upper limit of its safety threshold is automatically relaxed to 1.2 times the standard value. This dynamic adjustment avoids misclassifying normal seasonal variations as quality abnormalities. The adjustment coefficient is calculated based on a moving average algorithm, considering the median and interquartile range of the content distribution from the same period over the past five years.

[0105] The electronic signature system ensures the legal validity of the test results. The final report requires three levels of electronic signature confirmation: the testing operator's signature confirms the authenticity of the original data, the audit supervisor's signature acknowledges the compliance of the analysis methods, and the quality authorized person's signature approves the final conclusion. The signature process adopts a standardized workflow based on digital certificates, with hash value calculations covering the entire report and all attachments, and the signature timestamp is certified by an authoritative time service center.

[0106] Multi-site data comparison supports group-wide quality management. When an abnormality in ginsenosides is detected at a production base, the system automatically retrieves test results from other bases during the same period, finding similar fluctuations in parallel batches across the three bases. This group-level data analysis helps identify common problems, such as raw material quality issues from a unified supplier or deficiencies in shared process parameter settings. The comparison results are visualized as parallel coordinate graphs, highlighting the cross-site consistency of abnormal indicators.

[0107] Supply chain collaboration extends the boundaries of quality management. When an anomaly is confirmed due to raw material quality issues, the system generates a standardized supplier quality notification, automatically populating it with testing data, non-conformity descriptions, and corrective action requirements. The supplier portal allows users to view historical supply quality performance, including batch pass rates, key indicator fluctuation trends, and corrective action effectiveness scores. This collaborative mechanism promotes overall supply chain quality improvement.

[0108] Environmental monitoring data correlation analysis expands the dimensions of anomaly diagnosis. The system integrates temperature and humidity records from the production site, purified water quality data, and air cleanliness monitoring results. When abnormal components are detected, it automatically correlates with concurrent environmental parameters to identify potential influencing factors. For example, if the relative humidity in the extraction workshop corresponding to an abnormal batch of ginsenosides exceeds the control standard, this multi-factor correlation analysis improves the accuracy of root cause identification. Environmental data is displayed as overlaid time-series charts, supporting dynamic range selection and correlation analysis.

[0109] Automated document management improves the efficiency of quality record keeping. Drafts of subsequent documents triggered by abnormal events are automatically generated, including deviation handling records, change control applications, and CAPA reports. Intelligent data entry technology extracts key data from inspection results, maintaining consistency in document references. Controlled updates to document templates ensure compliance with the latest GMP document management standards. Electronic routing of approval processes supports multi-level countersigning and parallel review, significantly reducing document turnaround time.

[0110] The visualization and analysis tools provide a multi-faceted perspective on quality. The dashboard integrates various views: a geographic map displays the distribution of quality indicators for raw materials from different origins; box plots compare the process stability of each production line; scatter matrices analyze the correlation of key parameters; and a heatmap calendar presents the time-series clustering patterns of abnormal events. Interactive filtering supports drill-down analysis, allowing users to drill down from group-level aggregated data to specific batch production records. All views can be exported to standardized formats, facilitating the creation of monthly quality reports and regulatory briefings.

[0111] Example 5: The detection strategy optimization module dynamically adjusts the precision and response measures of the detection process based on the component risk level label. When the system identifies anomalies in low-risk components, it maintains standard detection parameters and only records deviation information for subsequent trend analysis. For medium-risk components, the flow rate parameter of the microfluidic chip is automatically lowered to the preset medium sensitivity mode, extending the chromatographic separation time to improve resolution, while the mass spectrometry scanning mode is switched to selective ion monitoring, focusing on fragment ions of specific mass numbers. High-risk components trigger a system-wide parameter upgrade, the microfluidic chip switches to a micro-flow rate level, the mass spectrometry uses a combined full-scan and multiple reaction monitoring mode, and the spectral acquisition integration time is extended by three times to improve the signal-to-noise ratio.

[0112] The medicinal herb traceability program immediately initiates a multi-level traceability query upon confirmation of high-risk components. The program first retrieves the incoming inspection records of all raw medicinal materials used in the current batch, verifying the consistency between the quality certificates provided by the supplier and internal testing results. It then traces back to the intermediate testing data of the preceding processes, analyzing the content trends of abnormal components at key process points such as extraction, purification, and concentration. Finally, it links to the planting or harvesting information of the raw medicinal materials, including original data such as growth environment parameters, harvesting time, and initial processing methods. The traceability process uses blockchain technology to store key evidence; each query generates a data packet containing a timestamp and digital signature, forming an immutable traceability chain.

[0113] The risk component interception system employs dual control measures: physical isolation and electronic tagging. Upon receiving an interception signal, the microfluidic fractionating device uses an array of solenoid valves to direct the target fraction to the isolation tank. Simultaneously, the electronic tag system updates, displaying the abnormal product as a red warning on the material management interface. The isolation tank is equipped with independent temperature monitoring and nitrogen protection systems to prevent sample deterioration during storage. An electronic lock mechanism restricts unauthorized personnel from accessing the isolated product, and all sampling operations require multi-level approval from the quality management department. Interception records automatically generate isolation notifications, detailing the detected values ​​of the abnormal component, the basis for the risk rating, and recommended retesting procedures.

[0114] The component risk rating module integrates multi-source information for comprehensive judgment. The system cross-references the current detection value of the abnormal component with three benchmarks: the content limit specified in the pharmacopoeia, the enterprise's internal control standard range, and the statistical distribution of historical detection data for the same product. Simultaneously, it queries the adverse reaction database to assess the types and frequencies of clinical reactions previously induced by the component at similar concentrations. The rating algorithm incorporates fuzzy logic to handle boundary cases; when multiple judgment dimensions conflict, it outputs a moderately conservative risk level and marks decision points requiring manual review.

[0115] The update mechanism of the medicinal herb traceability database enables positive feedback of abnormal information. Upon confirmation of raw material quality issues, the system automatically associates and stores the supplier code of that batch of medicinal materials with the characteristics of the abnormal components. When subsequent purchase orders are generated, the sampling ratio for raw materials from suppliers with historical problems is automatically increased. The database is simultaneously updated with the typical component spectrum model for that medicinal herb variety, adjusting the characteristic peak weight coefficients to improve sensitivity to similar anomalies. Long-term accumulated anomaly pattern data is used to optimize the supplier evaluation system, focusing on recurring problem types during quality audits.

[0116] The adaptive adjustment of detection parameters employs a smooth transition strategy. Flow rate changes are controlled using a ramp function to avoid chromatographic peak distortion caused by sudden flow abrupt changes. Mass spectrometry mode switching is performed when the ion transfer voltage crosses zero, minimizing signal interruption time. After adjusting the spectral acquisition parameters, the system automatically runs a rapid calibration program to verify the linear range of the detection sensitivity using standards. The entire switching process is completed within 30 seconds, with uninterrupted data acquisition during which only the parameter change time points are marked for subsequent analysis reference.

[0117] An electronic tagging system for risky components enables full lifecycle tracking. RFID tags are affixed to containers of isolated products, storing complete anomaly detection reports and handling procedures. The warehouse management system monitors the environmental conditions of the isolated area in real time, triggering alarms when conditions exceed preset limits. Electronic checkpoints are set up in the quality release process; products that have not undergone risk clearance approval cannot proceed to the next production stage. The final product disposal decision is recorded on the blockchain, ensuring the auditability of the disposal process.

[0118] A multi-system collaborative response constructs a complete quality protection network. Upon confirmation of a high-risk component, in addition to initiating product interception, the following coordinated measures are implemented simultaneously: the production management system suspends the scheduling of subsequent processes for that batch; the equipment management system locks relevant production line parameters to prevent misoperation; and the supply chain system suspends the approval of new orders for similar raw materials. This collaborative mechanism enables real-time communication between systems via an enterprise service bus, with all actions initialized within 60 seconds of risk confirmation.

[0119] Dynamic threshold management adapts to the seasonal fluctuations of medicinal materials. The system analyzes the distribution of testing data for specific components over the same period over the past three years and calculates monthly adjustment coefficients. For medicinal materials with concentrated harvest seasons, a wider acceptance range is set at the beginning of the market launch, gradually tightening the standards as storage time increases. The adjustment algorithm considers the correlation with climate data, automatically activating a stricter monitoring mode in years of extreme weather. Threshold updates are managed through version control, with each modification retaining approval records and scientific basis.

[0120] The electronic batch record system integrates a complete set of documents for anomaly handling. From initial detection and discovery, risk assessment, traceability investigation to final disposal, all key steps generate GMP-compliant electronic records. Hyperlinks are established between documents, supporting the tracing of original spectral data back from the final report. The signature process utilizes biometric technology, allowing operators to complete fingerprint or facial recognition confirmation on mobile devices. The audit trail function records the creation, modification, and approval timeline of documents, complying with 21 CFR Part 11 requirements.

[0121] The intelligent early warning system identifies potential risks in advance based on trend analysis. For components exhibiting fluctuations in the same direction across multiple consecutive batches, the system will issue a trend deviation warning even if a single test does not exceed limits. The warning rules consider the coordinated changes in process parameters, such as the expected range of change for a specific component after adjusting the extraction temperature. Warning information is categorized into observation level, attention level, and action level, corresponding to different response timeframes and approval levels. Trend analysis employs a moving average algorithm to eliminate random fluctuations and highlight meaningful systemic shifts.

[0122] The knowledge management system preserves best practices for handling anomalies. After each quality incident is closed, the system automatically extracts key elements to form a case study database, including anomaly description, root cause analysis, remedial measures, and verification results. Cases are categorized using multi-dimensional tags and support searching by ingredient name, risk level, process step, and other criteria. A similar case recommendation engine helps quality personnel quickly access historical experience; newly entered cases must be reviewed and confirmed by experts before publication.

[0123] The mobile quality dashboard enables real-time visualization of anomaly management. Production line supervisors' tablets display color-coded risk maps, intuitively showing the quality status of each batch. Clicking on an anomaly point opens a details page, including heatmap thumbnails, key value comparisons, and handling progress. Approval processes support mobile signing, allowing remote authorization of critical operations in emergencies. Dashboard data refreshes automatically every 15 seconds, ensuring on-site decisions are based on the latest test results.

[0124] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A precise detection system for liquor ingredients, characterized in that, Comprise: Multi-source spectral acquisition module: synchronously capture the near-infrared absorption spectrum sequence, Raman scattering spectrum time sequence characteristics and liquid chromatography mass spectrometry data stream of medicinal liquor samples, eliminate environmental light interference through adaptive filtering, and generate standardized spectral data matrix; Dynamic component modeling module: based on historical medicinal material spectrum library, construct a dynamic feature extraction network, input the standardized spectral data matrix into the feature extraction network, and output a theoretical component vector including a target active ingredient concentration interval; Multi-modal difference fusion module: multi-modal difference analysis is performed on the theoretical component vector and the reference component vector of the standard medicinal material database, the multi-modal difference analysis includes spectral absorption peak shift, characteristic peak intensity ratio deviation and chromatographic retention time distribution difference, and a multi-dimensional difference tensor of components is generated; Abnormal component positioning module: input the multi-dimensional difference tensor into a medicinal material correlation topological network, combine the medicinal material base source similarity matrix and the component interaction parameters, generate an abnormal component distribution probability heat map, and identify the abnormal component aggregation area; Detection strategy optimization module: according to the gradient distribution of the abnormal component distribution probability heat map, dynamically configure the detection parameters, including: enabling high-resolution microfluidic chip resampling for high-probability abnormal areas; Applying cross-modal spectral disturbance test to associated medicinal material components.

2. The liquor ingredient precision detection system according to claim 1, characterized in that, The multi-source spectral acquisition module specifically comprises: Heterogeneous sensor array: deploy a cooperative collection unit of ultraviolet-visible spectrophotometer, Fourier transform infrared spectrometer and mass spectrometer detector, synchronously acquire spectral absorption curve, Raman shift characteristic peak intensity distribution and molecular weight distribution chromatogram; Time sequence feature calibration unit: perform time domain alignment processing on the real-time collected spectral data stream, extract the peak shift rate and half peak width change parameters of the absorption spectrum, the displacement characteristic peak signal-to-noise ratio and phase angle distribution of the Raman spectrum, and the retention time gradient and peak area entropy value of the chromatographic peak; Noise suppression submodule: decompose the baseline shift noise caused by environmental temperature and humidity by wavelet packet transform, and eliminate high-frequency random interference by adaptive threshold filtering.

3. The liquor ingredient precision detection system according to claim 2, characterized in that, The dynamic component modeling module further comprises: Feature weighted fusion based on medicinal material compatibility rules, integrating the correlation characteristics of near-infrared spectrum, Raman fingerprint spectrum and mass spectrum fragments; Extracting multi-spectral cross-modal correlation feature vector through convolution attention mechanism; Output the theoretical component vector containing the medicinal material base source identification component, which is dynamically updated with medicinal material batches.

4. The liquor ingredient precision detection system according to claim 1, characterized in that, The multi-modal difference fusion module specifically comprises: Spectral shift calculation: sliding comparison of the absorption spectrum of the theoretical component vector and the reference component vector is performed with a preset wavelength window, the non-synchronous sampled spectral sequence is aligned by dynamic time warping algorithm, the cumulative shift of the characteristic peaks in each window is calculated, and a spectral shift vector is generated; Characteristic peak intensity evaluation: perform band energy decomposition on the Raman characteristic peaks of the theoretical vector and the reference vector, calculate the energy density ratio of the main peak and the secondary peak, extract the intensity deviation index of each displacement interval, and construct an intensity deviation vector; Chromatogram distribution difference detection: based on the structural similarity algorithm matching theory value and reference chromatographic peak time distribution, the phase error of retention time mutation point is calculated, the statistical dispersion of peak area distribution is quantified, and the chromatographic difference vector is generated; Multi-dimensional difference tensor generation: the spectral offset vector, intensity deviation vector and chromatographic difference vector are spliced into a three-dimensional tensor, and a three-order tensor with dimensions [component number x detection batch x difference type] is output through normalized processing with feature saliency weighting.

5. The liquor ingredient precision detection system according to claim 4, characterized in that, The abnormal component positioning module specifically includes: Medicinal material topology modeling: constructing a component correlation topology graph based on medicinal material base information, labeling the chemical bond connection relationship and molecular force parameters between components, superimposing solvent effect constraint conditions of liquor preparation process in the topology graph, and generating a chemical topology model including a molecular similarity matrix and a component affinity matrix; Abnormal propagation simulation: mapping the multi-dimensional difference tensor to the corresponding nodes of the chemical topology model, and performing abnormal component diffusion deduction based on a graph neural network, which includes: calculating the abnormal component migration attenuation factor according to the molecular force parameters; capturing cross-component abnormal correlation characteristics through a multi-head attention mechanism; and simulating the diffusion path of the abnormal component in the topology network using the Monte Carlo method; Heat map generation: statistically analyzing the abnormal occurrence frequency of each chemical component in the simulated diffusion, calculating the abnormal component residence probability value based on the molecular affinity parameter, and generating an abnormal probability heat map covering the entire component set; Physical component positioning: performing spatial clustering analysis on the abnormal probability heat map to identify a set of chemical components with abnormal probability exceeding a preset threshold.

6. The precise detection system for liquor ingredients according to claim 5, characterized in that, The detection strategy optimization module specifically includes: High-resolution resampling trigger unit: when the probability value of a specific area in the abnormal probability heat map exceeds a first threshold, an instruction is issued to the microfluidic chip corresponding to the area to perform: increasing the spectral sampling resolution to 5 times the original value; simultaneously enabling the mass spectrometry multiple reaction monitoring mode to capture the characteristic ion fragment intensity ratio; and deploying a transient peak shape recorder at the detection end to store abnormal chromatographic peak fragments; Spectral perturbation test unit: applying multi-band spectral excitation to the associated chemical components with the largest gradient change in the abnormal probability heat map.

7. The liquor ingredient precision detection system according to claim 6, characterized in that, The abnormal component positioning module also outputs a suspicious component identification set and an abnormal propagation main path; The suspicious component identification set binds the chemical components with the medicinal material base based on the clustering results of the abnormal probability heat map; The abnormal propagation main path is generated by recording the highest frequency component migration sequence in the Monte Carlo simulation. 8.The system of claim 1, wherein, Further includes: Component risk rating module: calculating the abnormal confidence of each chemical component based on the distribution gradient of the abnormal probability heat map; Matching the abnormal confidence and the safety threshold based on the historical adverse reaction database to generate a component risk level label; Detection report generation module: integrating the component risk level label, the abnormal component distribution probability heat map, and the multi-dimensional difference tensor to output a structured detection report.

9. The liquor ingredient precision detection system according to claim 8, characterized in that, The detection strategy optimization module further includes: Dynamically adjusting the flow rate parameters of the microfluidic chip according to the component risk level label; When a high-risk component is detected, automatically trigger the execution of the medicinal material base traceability program.

10. The system according to claim 8, characterized in that, The ingredient risk rating module further comprises performing: When an abnormal chemical component whose confidence exceeds a safety threshold is detected, a risk ingredient interception signal is generated; based on the risk ingredient interception signal, a physical isolation operation is triggered by an automatic fractionation device, and an abnormal ingredient batch mark of a medicinal material base source traceability database is synchronously updated.

Images