Propolis liquid quality detection data processing method and system

Abstract

The invention discloses a propolis liquid quality detection data processing method and system, and relates to the technical field of data processing.The method comprises the steps that spectral data of a batch of propolis liquid to be detected is obtained, the content of symbolic components is analyzed, and production stage parameters are obtained; the technical key points are as follows: dynamic correction of a threshold value is realized through simulation feedback, a balance point is found between detection accuracy and efficiency, on one hand, a correction coefficient and feedback rate comparison mechanism is introduced, and whether threshold value correction needs to be performed or not is intelligently judged according to a simulation feedback result; the problem that the detection efficiency is reduced due to redundant correction actions is effectively avoided, and simplification and high efficiency of the detection process are ensured; and on the other hand, when the correction feedback displays a better result, secondary correction can be accurately performed on the adaptive threshold value, and the adaptive adjustment mechanism enables the threshold value setting to be closer to the detection requirement, so that the influence of parameter adjustment on the adaptive threshold value is reflected more comprehensively, and the detection accuracy is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically to a method and system for processing propolis liquid quality detection data. Background Art

[0002] The data processing process in propolis liquid quality detection includes data collection, preprocessing, analysis, calculation of quality reference indicators, and comparison decision-making; firstly, spectral data of propolis liquid is collected through ultraviolet-visible spectroscopy instruments or high performance liquid chromatography (HPLC) to obtain information on key components such as phenols, flavonoids, and amino acids; subsequently, a baseline correction algorithm is used to preprocess the spectral data to remove the interference of impurities and background signals; then, the fuzzy C-means clustering algorithm (FCM) is used to perform clustering analysis on the preprocessed data, dividing the spectral data into different clustering clusters, and each cluster represents the spectral characteristics of a component; according to the clustering results, quality reference indicators of each cluster are calculated, such as the height difference and area difference of absorption peak comparison; finally, the quality reference indicators are compared with a preset threshold, and if it exceeds the threshold, it is determined that the quality of the propolis liquid does not meet the standard, otherwise it meets the standard; this process ensures the accuracy and reliability of the detection, but the setting of its threshold usually adopts a fixed setting method and lacks flexibility.

[0003] During the process of processing propolis liquid quality detection data, traditional quality control methods usually adopt fixed thresholds and cannot adapt to the quality change trend, resulting in an increased risk of misjudgment and missed judgment; in addition, the setting of thresholds in traditional methods often mostly relies on subjective experience and lacks objectivity and scientificity; although the threshold determination and setting can also adopt an adaptive adjustment method, the effect achieved after the adjustment is not always satisfactory. For example, there is still a certain false alarm rate for the threshold obtained after adaptive adjustment. If you want to reduce this part of the false alarm rate, you need to continue to perform repeated adjustment operations, and the detection efficiency cannot be guaranteed, and a certain balance cannot be achieved between the accuracy and efficiency of the detection. Summary of the Invention

[0004] (I) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides a method and system for processing propolis liquid quality detection data. By running this method, the problems raised in the background art are solved.

[0005] (II) Technical Solutions To achieve the above objectives, the present invention is realized through the following technical solutions: A method for processing propolis liquid quality detection data, and the steps of this method are as follows: Obtaining spectral data of the batch of propolis liquid to be tested, analyzing the content of characteristic components, and obtaining production stage parameters; preprocessing the spectral data; extracting spectral features based on the preprocessed spectral data, and constructing a multidimensional vector matrix; and also including: Based on the multi-dimensional vector matrix, the SVM classifier is used to make qualified judgments; based on the extracted spectral features, the isolation forest is used to make abnormal judgments; the judgment results are summarized, and based on the pre-built rule engine, the compliance rate of the current batch is obtained, and compared with the compliance rate of the previous batch, the quality change trend is analyzed; Construct an adaptive threshold model. On the premise of determining the initial threshold, make decisions to adjust the initial threshold according to the quality change trend results to obtain the adaptive thresholds corresponding to different types of characteristic components. Run the constructed production process digital twin and analyze the production stage parameters to determine whether to trigger the correction mechanism. Introduce the correction coefficient into the virtual model in the digital twin for dynamic compensation. Compare the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, and compare the results in turn to determine the final threshold. In the actual detection process, a comparative analysis is performed based on the final threshold and a decision is made whether to issue a warning signal.

[0006] Furthermore, the characteristic ingredient types include at least: flavonoids and terpenes; the production stage parameters include at least: the adjustment direction and magnitude of the temperature, pH value and feed rate of the propolis liquid batch to be tested during the production stage.

[0007] Further, the spectral data is preprocessed, including at least: spectral noise reduction and baseline correction; Spectral feature extraction: The characteristic peak positioning algorithm was used to detect the characteristic peaks corresponding to flavonoids and terpenes, and the peak height ratio was calculated; the number of dimensions in the multidimensional vector matrix was consistent with the number of characteristic component types.

[0008] Furthermore, after summarizing the judgment results, the process of running the rule engine is as follows: Screening the propolis liquid samples that meet the category label as qualified and normal from the batches of propolis liquid to be tested, and counting the proportion of the number of screened samples to the total number of propolis liquid in the batches of propolis liquid to be tested to obtain the compliance rate; When compared with the compliance rate of the previous batch of propolis liquid: If the compliance rate of the previous batch of propolis liquid exceeds the compliance rate of the batch of propolis liquid to be tested, the quality change trend is judged to be a downward trend; otherwise, the quality change trend is an upward trend.

[0009] Furthermore, the initial threshold is μ±s×σ; wherein μ and σ represent the mean and standard deviation corresponding to the same type of characteristic components, respectively; the value range of s is: s>0.

[0010] Further, when the quality change trend is an upward trend, the formula on which the dynamic adjustment model is based is: ; Meanings of letters: T_new: Adaptive threshold, w1: Weight 1, T_old: Initial threshold, w2: Weight 2, and 1 > w1 > w2 > 0, μ_current: Mean of the batch to be detected; When the quality change trend is a downward trend, the formula on which the dynamic adjustment model is based is: ; ; Meanings of letters: T_new_max: Upper limit value of the adaptive threshold, T_old_max: Upper limit value of the initial threshold, k: Expansion factor, and k > 0, T_new_min: Lower limit value of the adaptive threshold, T_old_min: Lower limit value of the initial threshold, σ_current: Mean of the batch to be detected.

[0011] Further, when adjusting the parameters of any production stage, the correction mechanism is triggered; In the correction mechanism, the correction coefficient introduced by dynamic compensation is obtained by running a pre - constructed correction function model: Collect the mean μ1 and standard deviation σ1 of the contents of different types of signature components in the previous several batches of propolis liquid, calculate the difference Δμ between the mean μ1 and μ_current for the same type of signature component, and the difference Δσ between the standard deviation σ1 and σ_current; Among them, both Δμ and Δσ are greater than 0; Obtain the corresponding correction coefficient Q for the batch to be detected, and perform the Sigmoid function correction action based on Δμ and Δσ; The corrected adaptive threshold is Q × T_new.

[0012] Further, in comparing the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, the comparison process is as follows: When the feedback rates corresponding to the corrected adaptive threshold all exceed the feedback rate corresponding to the original adaptive threshold, maintain the original adaptive threshold and use it as the final threshold; When the feedback rate corresponding to any corrected adaptive threshold does not exceed the feedback rate corresponding to the original adaptive threshold, call the corresponding corrected adaptive threshold and use it as the final threshold; When the feedback rates corresponding to the corrected adaptive thresholds do not exceed the feedback rates corresponding to the original adaptive threshold, compare the feedback rates corresponding to the two corrected adaptive thresholds, and select the one with the smaller feedback rate as the final threshold; Among them, the feedback rate is either the false alarm rate or the missed alarm rate.

[0013] Furthermore, the process of comparative analysis based on the final threshold is as follows: Compare the signature component under any number in the propolis liquid of the batch to be detected with its corresponding final threshold: When a single type of signature component exceeds the standard, a first-level warning signal is issued; When multiple types of signature components exceed the standard, a second-level warning signal is issued.

[0014] A data processing system for propolis liquid quality detection, the system includes: Data acquisition and processing module: Obtain the spectral data of the propolis liquid of the batch to be detected, analyze the content of signature components, and obtain production stage parameters; preprocess the spectral data; Feature engineering construction module: Extract spectral features based on the preprocessed spectral data, and construct a multi-dimensional vector matrix; Multi-model fusion analysis module: According to the multi-dimensional vector matrix, use the SVM classifier to judge compliance; according to the extracted spectral features, use the isolation forest to judge anomalies; summarize the judgment results, and based on the pre-built rule engine, obtain the pass rate of the current batch, and compare it with the pass rate of the previous batch to analyze and obtain the quality change trend; Dynamic threshold decision module: Construct an adaptive threshold model. On the premise of determining the initial threshold, make a decision to adjust the initial threshold according to the quality change trend result to obtain the adaptive thresholds corresponding to different types of signature components; Digital twin construction module: Run the constructed digital twin of the production process, and analyze the production stage parameters to judge whether to trigger the correction mechanism, and introduce a correction coefficient in the virtual model in the digital twin for dynamic compensation; compare the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, and determine the final threshold according to the comparison result; Result output and warning module: In the actual detection process, conduct comparative analysis based on the final threshold and decide whether to issue a warning signal.

[0015] (III) Beneficial effects The present invention provides a method and system for processing propolis liquid quality detection data, having the following beneficial effects: Comprehensively consider multi-dimensional information: Using the RBF kernel function, it can handle non-linear classification problems, make qualified judgments according to preset component standards, and has a high classification accuracy; adopting the Isolation Forest algorithm can detect whether the peak height ratio deviates from the normal range to identify abnormal samples, and has high sensitivity to outliers. By combining the analysis results of SVM and the Isolation Forest, the advantages of multiple models can be integrated to improve the accuracy and robustness of the overall analysis; even if a certain model performs poorly in some cases, the analysis results of other models can still provide effective information; Improve the accuracy and reliability of detection: Dynamic threshold decision-making can adjust the threshold in real time according to different quality change trends to ensure that the threshold always matches the actual quality level; when the quality improves, dynamic threshold decision-making strictly screens qualified products by narrowing the qualified range, thereby improving the accuracy of detection; when the quality deteriorates, dynamic threshold decision-making avoids over-eliminating qualified products by expanding the qualified range to accommodate larger quality fluctuations, thereby improving the reliability of detection; Balance the accuracy and efficiency of detection: Achieve dynamic correction of the threshold through simulation feedback, and find a balance between the accuracy and efficiency of detection. On the one hand, introduce a correction coefficient and a feedback rate comparison mechanism, and intelligently judge whether threshold correction is required according to the simulation feedback results, effectively avoiding the problem of decreased detection efficiency caused by redundant correction actions, and ensuring the simplicity and high efficiency of the detection process; on the other hand, when the correction feedback shows a better result, it can accurately perform secondary correction on the adaptive threshold. This adaptive adjustment mechanism makes the threshold setting closer to the detection requirements, thus more comprehensively reflecting the impact of parameter adjustment on the adaptive threshold and further improving the accuracy of detection; in addition, it also supports real-time remote monitoring and early warning, can timely detect quality fluctuations and send out early warning signals, and improves the flexibility and practicality of quality control. Brief Description of the Drawings

[0016] Figure 1 It is a schematic diagram of the overall method flow of a method for processing propolis liquid quality detection data in the present invention. Detailed Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1: Please refer to Figure 1 , this embodiment provides a method for processing propolis liquid quality detection data, and the method includes: S1. Multi-source data collection Obtain the spectral data of the propolis liquid in the batch to be detected, analyze the content of the marker components, and obtain the production stage parameters; among them, the types of marker components include at least flavonoids and terpenoids; The propolis liquid in the batch to be detected contains several propolis liquid samples with different numbers; for example: the propolis liquid in the batch to be detected contains 12 propolis liquid samples numbered B01, B02,... up to B12; The spectral data is collected by a spectrometer: Specifically, an Ocean Optics QE Pro spectrometer can be used; parameter settings: integration time 8 ms, averaging times 10 times; output example: wavelength array [200.0, 200.5,..., 800.0] nm, corresponding intensity values [0.12, 0.45,..., 0.03]; The content of the marker components is analyzed by an HPLC chromatograph: Specifically, an Agilent 1260 Infinity II liquid chromatograph can be used; gradient elution program: 30% acetonitrile from 0 to 5 min, 70% acetonitrile from 5 to 15 min; detection wavelength: set dual-wavelength detection, such as 275 nm for flavonoids and 325 nm for terpenoids; output example: flavonoid content 6.85%, terpene content 2.32%; The production stage parameters at least include: the adjustment direction and magnitude of the temperature, pH value, and feed rate of the propolis liquid in the batch to be detected during the production stage; for example: the temperature adjustment increases, and the increase amount is 2 °C; It should be noted that in this embodiment, the marker components are: flavonoids and terpenoids.

[0019] S2. Data preprocessing Preprocess the spectral data, including at least spectral noise reduction and baseline correction; Among them, spectral noise reduction: adopt Savitzky-Golay filtering (window width 9, polynomial order 3); Savitzky-Golay filtering (function: filter clutter with an algorithm to make the spectral data cleaner; for example, the originally jumping data points become smoother after processing): Window width = 9 (covering 4.5 nm), polynomial order = 3; Example: Spectrum before processing: [0.12, 0.15, 0.21, 0.18, 0.16,...]; Spectrum after processing: [0.14, 0.16, 0.20, 0.19, 0.17,...]; Effect: The signal-to-noise ratio is increased from 15 dB to 32 dB; Baseline correction: The Asymmetric Least Squares algorithm eliminates scattering interference; Example: Original baseline drift: 0.05 (maximum); Corrected baseline: 0.002 (maximum); Absorbance value before correction: 2.15 → 1.98 after correction.

[0020] S3. Feature engineering Based on the preprocessed spectral data, spectral feature extraction is performed, and a multi-dimensional vector matrix is constructed according to the content of the signature components; among them, the specific number of dimensions of the multi-dimension is the same as the number of types of signature components; Spectral feature extraction: The feature peak localization algorithm is used to detect the feature peaks corresponding to flavonoids and terpenoids, and the peak height ratio is calculated; among them, the feature peak localization algorithm: find the zero-crossing point of the second derivative; Example: Feature peaks at 275 nm (flavonoids) and 325 nm (terpenoids) are detected; Calculate the peak height ratio: 275 nm / 325 nm = 1.82 (the normal range is 1.5 to 2.0); The constructed multi-dimensional vector matrix is specifically: Two-component matrix: Construct a two-dimensional feature vector: v = [C 黄酮类 , C 萜烯类 ; Example data is as follows in the table: Table 1: Content of signature components under different numbers in the same batch:

[0021] Specifically, the specific number of dimensions of the multi-dimension is 2, and the number of types of signature components is also 2, which are flavonoids and terpenoids respectively; Hybrid modeling optimization: Compress the original multi-dimensional (e.g., 12-dimensional) component space into a 2-dimensional core component space, improving the model training efficiency. Although it will cause a decrease in the accuracy of subsequent threshold setting, the subsequent adaptive adjustment + threshold correction can make up for the problem of reduced accuracy.

[0022] S4. Multi-model fusion analysis Based on the multi-dimensional vector matrix, use the SVM classifier to judge the qualification of the propolis liquid to be detected; Based on the extracted spectral features, use the Isolation Forest to judge the abnormality of the propolis liquid to be detected; Summarize the judgment results, and based on the pre-built rule engine, obtain the pass rate of the current propolis liquid to be detected. Compared with the pass rate of the previous batch of propolis liquid, analyze the quality change trend; Static detection layer adjustment: SVM classifier: Use the RBF kernel function, C = 1.5; Table 2: Example training set:

[0023] As can be seen from Table 2 above: Only when both flavonoids and terpenoids are qualified can it be indicated that the category label of the propolis liquid sample under the corresponding number is qualified; conversely, if any one category is unqualified, it means that the category label of the propolis liquid sample under the corresponding number is unqualified; the SVM classifier makes a qualification judgment according to the preset component standards (such as the content ranges of flavonoids and terpenoids). Isolation forest algorithm: Input features: The peak height ratio obtained using the feature peak localization algorithm; Detection target: Detect whether the peak height ratio deviates from the normal range to identify abnormal samples; Example: Sample data: For the sample numbered B02, the peak height ratio = 2.2 (exceeding the normal range of 2.0); Detection process: Peak height ratio 2.2 > 2.0 → Abnormal; Detection result: The sample numbered B02 is abnormal (if the peak height ratio = 1.2, the sample is normal); After summarizing the judgment results, the process of running the rules engine is as follows: Screen out the propolis liquid samples that are both qualified in category label and normal from the propolis liquid to be detected, and count the proportion of the number of screened samples in the total number of the propolis liquid to be detected to obtain the pass rate; When comparing with the pass rate of the previous batch of propolis liquid: If the pass rate of the previous batch of propolis liquid exceeds the pass rate of the propolis liquid to be detected, the quality change trend is a downward trend; if the pass rate of the previous batch of propolis liquid does not exceed the pass rate of the propolis liquid to be detected, the quality change trend is an upward trend (the same pass rate is also classified as an upward trend because the probability of remaining the same is less than 0.01%); Regarding the solution in S4, the following explanations are provided: I. Reasons for not directly using the mean μ and standard deviation σ to judge the quality change trend: As statistical values, the mean and standard deviation have a certain role in measuring the data dispersion degree, but they have the following limitations, making them not suitable for directly judging the quality change trend; Excessive information simplification: The mean and standard deviation are simplifications and generalizations of data, which may ignore some important detailed information, such as outliers and distribution patterns; in quality analysis, outliers may exactly reflect key quality problems, while the mean and standard deviation cannot capture these details; II. Reasons for not using the LSTM network model for prediction: Although the LSTM model can handle long-term dependencies, capture non-linear relationships, and has a certain robustness to outliers, it also has the following deficiencies in quality change trend prediction: High data demand: The LSTM model requires a large dataset to support it in capturing complex patterns in the data. In quality analysis, if the amount of data is limited, the prediction effect of the LSTM model may not be good; Lack of interpretability: As a deep learning model, the prediction results of the LSTM model are often difficult to interpret. In quality analysis, users not only care about the prediction results but also hope to understand the basis and reasons for the prediction in order to carry out subsequent quality improvements; III. Reasons and advantages of adopting multi-model fusion analysis and rule engine: Integrating the advantages of multiple models: SVM classifier: Using the RBF kernel function, it can handle non-linear classification problems and make qualified judgments according to preset component criteria (such as the content ranges of flavonoids and terpenoids), with a high classification accuracy; Isolation Forest algorithm: It can detect whether the peak height ratio deviates from the normal range to identify abnormal samples and has a high sensitivity to outliers; Multi-model fusion: By combining the analysis results of SVM and Isolation Forest, the advantages of the two models can be integrated to improve the accuracy and robustness of the overall analysis. Even if a certain model performs poorly in some cases, the analysis results of other models can still provide effective information; Rule engine improves decision-making efficiency: Clear rules: The rule engine makes decisions according to preset rules (such as pass rate calculation rules, quality change trend judgment rules), improving the transparency and interpretability of decision-making; Efficient decision-making: The rule engine can process and analyze data in real time and quickly give decision results, improving decision-making efficiency; Comprehensively considering multi-dimensional information: Multi-dimensional vector matrix: The SVM classifier uses the multi-dimensional vector matrix to make qualified judgments, considering the comprehensive influence of multiple component indicators; Spectral features: The Isolation Forest algorithm uses the extracted spectral features to make abnormal judgments, capturing the spectral characteristics of the samples; Comprehensive judgment: By summarizing the analysis results of the two models, the rule engine can comprehensively consider multiple-dimensional information of the samples, improving the comprehensiveness and accuracy of the analysis; In summary, adopting multi-model fusion analysis and rule engine for quality change trend analysis has advantages such as integrating the advantages of multiple models, improving decision-making efficiency, and comprehensively considering multi-dimensional information. At the same time, it also has other advantages such as presenting innovative analysis methods, improving analysis robustness, and providing new perspectives and solutions. Therefore, this method has great application value and promotion prospects in quality analysis.

[0024] S5. Dynamic threshold decision Build an adaptive threshold model. On the premise of determining the initial threshold μ ± s×σ, make decisions to adjust the initial threshold μ ± s×σ according to the quality change trend results, so as to obtain the adaptive thresholds corresponding to different types of signature components; Among them, the mean μ and the standard deviation σ represent the mean and the standard deviation corresponding to the same type of signature component; The value range of s is: s > 0; The example is as follows: If the initial threshold: μ ± 3σ (based on the data of the first 3 batches); calculate: Flavonoids: mean μ = 4.02%, σ = 0.25 → Initial threshold: 3.27% to 4.77%; Terpenoids: mean μ = 1.65%, σ = 0.18 → Initial threshold: 1.11% to 2.19%; Make decisions to adjust the initial threshold μ ± s×σ according to the quality change trend results: When the quality change trend is an upward trend, indicating that the quality gets better, the formula based on which the dynamic adjustment model is: ; Meaning of letters: T_new: Adaptive threshold (new threshold); w1: Weight 1, representing the weight of the initial threshold (old threshold); T_old: Initial threshold (old threshold); At this time, T_old = μ ± s×σ; w2: Weight 2, representing the weight of the current batch mean; and 1 > w1 > w2 > 0; μ_current: Mean of the current batch (batch to be detected); The weight coefficients are determined by the coefficient of variation method. The coefficient of variation method is a method of assigning weights to each evaluation index according to the degree of variation between the current value and the target value of each evaluation index; if the values of a certain index vary greatly and can clearly distinguish each evaluated object, it indicates that the resolution information of this index is rich, so a larger weight should be given to this index; conversely, if the values of each evaluated object on a certain index vary little, then the ability of this index to distinguish each evaluation object is weak, so a smaller weight should be given to this index; this method directly utilizes the information contained in each index and calculates the weights of the indexes through calculation, so it has objectivity; When the quality change trend is a downward trend, indicating that the quality gets worse, the formula based on which the dynamic adjustment model is: ; ; Meaning of letters: T_new_max: Upper limit value of the adaptive threshold; T_old_max: Initial threshold upper limit value; k: Expansion factor, used to expand the new threshold range, and the value range is: k > 0 (the value is 1 in this embodiment); T_new_min: Adaptive threshold lower limit value; T_old_min: Initial threshold lower limit value; σ_current: Standard deviation of the current batch (batch to be detected); The influence of units and percentage signs is not considered in the formula; In summary, when running the formula in the dynamic adjustment model, different formulas are adopted according to different conditions (the quality change trend is an upward or downward trend) to calculate the required adaptive threshold T_new; Design logic or reason of the dynamic adjustment model: Mainly based on the real-time feedback of the quality change trend and the need for adaptive adjustment, the model can flexibly adjust the threshold range under different quality trends to make it more in line with the actual quality level; It should be noted that (usually the quality of adjacent batch products is similar or the difference is small): 1. Design purpose of weight 1 being greater than weight 2: Maintaining stability: Weight 1 (weight of the initial threshold) is greater than weight 2 (weight of the current batch mean), ensuring that the new threshold will not change violently due to the fluctuation of the current batch data and maintaining the stability of threshold adjustment; Smooth transition: When the quality trend changes, the new threshold can make a smooth transition to avoid the incoherence of the quality control strategy caused by mutations; 2. Adjustment logic when the quality trend improves: Narrowing the threshold range: When the quality trend improves, the current batch mean is closer to the ideal quality level. The model increases the weight of the current batch mean, making the new threshold range shrink towards the mean and narrowing the qualified range; Purpose: When the quality is improving, more strictly control the quality of propolis liquid to avoid missed inspections and ensure the stability and consistency of product quality; 3. Adjustment logic when the quality trend deteriorates: Expanding the threshold range: When the quality trend deteriorates, the current batch data may contain greater fluctuations. The model introduces an expansion factor to expand the qualified range; Purpose: When the quality is declining (there are many qualified products misjudged as unqualified), be more tolerant of quality fluctuations, avoid over-eliminating qualified products, and at the same time remind the enterprise to pay attention to the declining quality trend and take timely measures for improvement; First, from the perspective of the quality loss principle, when the distribution curve of quality characteristic data exceeds the specified standard range, the proportion of non-conforming products will increase; in the case of quality decline, appropriately expanding the qualified range can include more products that were originally on the verge of being qualified into the qualified range, thereby reducing the number of non-conforming products and the quality loss; Second, from the perspective of quality cost management, quality decline will lead to an increase in internal loss costs (such as rework, etc.) and external loss costs (such as loss of reputation, etc.); by appropriately expanding the qualified range, enterprises can reduce the number of non-conforming products and lower internal loss costs; at the same time, this can also avoid external loss costs caused by excessive elimination of qualified products; In summary, when quality declines, the design logic of expanding the qualified range is based on a comprehensive consideration of the quality loss principle and quality cost management; by appropriately expanding the qualified range, enterprises can avoid excessive elimination of qualified products, reduce the quality loss cost, and at the same time remind enterprises to pay attention to the trend of quality decline and take timely measures for improvement; this design not only ensures the flexibility of quality control but also takes into account the economic benefits of enterprises and market demands; Products (propolis liquid) under different quality change trend conditions can be marked separately for convenient subsequent management; Taking the flavonoid components and terpene components provided by users as an example, the initial thresholds are both μ ± 3σ: Flavonoids: mean μ = 4.02%, σ = 0.25 → initial threshold: 3.27% to 4.77%; Terpenoids: mean μ = 1.65%, σ = 0.18 → initial threshold: 1.11% to 2.19%; Suppose the data of the current batch is as follows: Flavonoids: μ_current = 4.2%, σ_current = 0.20; Terpenoids: μ_current = 1.5%, σ_current = 0.22; Set weight 1 = 0.7, weight 2 = 0.3, k = 1; Example 1: Quality trend gets better Flavonoids: New threshold lower limit: 0.7×3.27% + 0.3×4.2% = 3.549%; New threshold upper limit: 0.7×4.77% + 0.3×4.2% = 4.599%; New threshold range: 3.549% to 4.599% (narrower than the initial threshold range); Terpenoids: New threshold lower limit: 0.7×1.11% + 0.3×1.5% = 1.227%; New upper threshold: 0.7×2.19% + 0.3×1.5% = 1.983%; New threshold range: 1.227% to 1.983% (narrower than the initial threshold range); Effect: The new threshold range is narrowed, which is more in line with the actual level after quality improvement, and strictly screens qualified products; Example 2: The quality trend deteriorates Flavonoids: New lower threshold: 3.27 - 1×0.2 = 3.07%; New upper threshold: 4.77 + 1×0.2 = 4.97%; New threshold range: 3.07% to 4.97% (wider than the initial threshold range); Terpenoids: New lower threshold: 1.11 - 1×0.22 = 0.89%; New upper threshold: 2.19 + 1×0.22 = 2.41%; New threshold range: 0.89% to 2.41% (wider than the initial threshold range); Effect: The new threshold range is widened, accommodating greater quality fluctuations, avoiding over-eliminating qualified products, and at the same time prompting enterprises to pay attention to the trend of quality decline; Specifically, the technical effects achieved by the above solutions are: Improve the accuracy and reliability of detection: Dynamic threshold decision-making can adjust the threshold in real time according to different quality change trends, ensuring that the threshold always matches the actual quality level; when the quality improves, dynamic threshold decision-making strictly screens qualified products by narrowing the qualified range, thereby improving the accuracy of detection; when the quality declines, dynamic threshold decision-making accommodates greater quality fluctuations by widening the qualified range, avoiding over-eliminating qualified products, thereby improving the reliability of detection; Enhance the efficiency of the model: The dynamic adjustment model can adapt to different quality change trends and adjust the threshold without manual intervention, improving the efficiency of the model; Smooth the threshold adjustment process: Dynamic threshold decision-making realizes the smooth transition of threshold adjustment by introducing weight coefficients and expansion factors, avoiding the incoherence of quality control strategies caused by mutations.

[0025] S6. Result output and warning During the actual detection process, compare the signature components under any number in the batch of propolis liquid to be detected with the adaptive threshold; if a single type of signature component exceeds the standard, a first-level warning signal is issued; if multiple types of signature components exceed the standard, a second-level warning signal is issued.

[0026] Example 2: Based on Example 1, in the subsequent steps of the S5 dynamic threshold decision in Example 1, S5.1, dynamic cognitive digital twin (correcting the threshold, simulating feedback), is added; Run the constructed digital twin of the production process in real time, and analyze the production stage parameters to determine whether to trigger the correction mechanism, and introduce a correction coefficient in the virtual model within the digital twin for dynamic compensation; Compare the feedback rate obtained by simulation under the corrected adaptive threshold with that under the original adaptive threshold; When the feedback rates corresponding to all the corrected adaptive thresholds exceed the feedback rate corresponding to the original adaptive threshold, maintain the original adaptive threshold and use it as the final threshold; When the feedback rate corresponding to any one of the corrected adaptive thresholds does not exceed the feedback rate corresponding to the original adaptive threshold, call the corresponding corrected adaptive threshold and use it as the final threshold; When the feedback rates corresponding to all the corrected adaptive thresholds do not exceed the feedback rate corresponding to the original adaptive threshold, compare the feedback rates corresponding to the two corrected adaptive thresholds, and select the one with the smaller feedback rate as the final threshold; Among them, the feedback rate is either the false alarm rate or the miss rate; Dynamic mapping relationship: The digital twin is driven by real-time data (production stage parameters) to establish a two-way closed loop between the physical system and the virtual model; when the production stage parameters (such as temperature, pH value, feed rate) are adjusted in the production stage, the virtual model will immediately respond and trigger the correction mechanism, and perform simulation comparison through the virtual model within the digital twin; Analyze production stage parameters: Analyze whether any production stage parameter is adjusted. If it is adjusted, trigger the correction mechanism; Operation of the correction function model: Collect the mean μ1 and standard deviation σ1 of the contents of different types of signature components in the previous several batches of propolis liquid, calculate the difference Δμ between the mean μ1 and μ_current for the same type of signature component, and the difference Δσ between the standard deviation σ1 and σ_current; Among them, both Δμ and Δσ are greater than 0; When obtaining the corresponding correction coefficient Q for the current batch, perform the Sigmoid function correction action, and the formula on which this action is based is as follows: ; Meaning of letters: S(): Sigmoid function; S(x) = 1 / (1 + e^(-x)); where x represents (a×Δμ + b×Δσ); a and b: Both are adjustment parameters used to control the shape of the Sigmoid function, thereby adjusting the sensitivity of the correction coefficient Q to changes in Δμ and Δσ; among them, the value ranges of a and b are both greater than 0; Design logic of the running formula in the correction function model: Range constraint: To ensure that the correction coefficient Q does not make too large an adjustment to the original adaptive threshold (so as not to affect the accuracy), we use the Sigmoid function S(x) to map the linear combination a×Δμ + b×Δσ to between 0 and 1; the constant terms (1 and 0.05) in the formula are used to ensure that the correction amplitude is neither too large nor too small; Sensitivity control: The adjustment parameters a and b are used to adjust the steepness of the Sigmoid function; larger values of a and b will make the Sigmoid function steeper, thereby increasing the sensitivity of the correction coefficient Q to changes in Δμ and Δσ; smaller values of a and b will make the Sigmoid function flatter, reducing the sensitivity of the correction coefficient Q; Comprehensive reflection: The correction coefficient Q takes into account the changes in both the mean μ and the standard deviation σ, and can more comprehensively reflect the impact of parameter adjustment on the adaptive threshold; The corrected adaptive threshold is Q×T_new; Since the final value of Q has 2, there are also two sets of range values corresponding to the corrected adaptive threshold; Assume that for the first 100 batches, the mean of flavonoids μ1 = 5.0% and the standard deviation σ1 = 0.5; Taking flavonoids as an example only: μ_current = 4.2%, σ_current = 0.20; Let a = 0.5, b = 0.5; T_new is: from 3.549% to 4.599%; 1. Calculate the change amounts of the mean and the standard deviation: Δμ = 5.0% - 4.2% = 0.8%; Δσ = 0.5 - 0.20 = 0.3; 2. Calculate the correction coefficient: Q = 1 ± 0.05×(1 / (1 + e^(-(0.5×0.8 + 0.5×0.3)))) = 1 ± 0.032; Therefore, the values of Q are 1.031 and 0.968; 3. Calculate the corrected adaptive threshold: When Q = 1.032, Q×T_new → the corrected adaptive threshold: from 3.663% to 4.746%; When Q = 0.968, Q×T_new → the corrected adaptive threshold: from 3.435% to 4.452%; It should be noted that for the calculation of the threshold, only the last two decimal places can be retained as needed; 4. Comparison of simulated false alarm rates: The false alarm rate of the original adaptive threshold: 0.12%; The false alarm rates of the corrected adaptive thresholds: 0.1167% and 0.1233%; 5. Determine the final threshold: Meet the following conditions: When the feedback rate (0.1167%) corresponding to any one of the corrected adaptive thresholds does not exceed the feedback rate (0.12%) corresponding to the original adaptive threshold, then call the corresponding corrected adaptive threshold (3.663% to 4.746%) and use it as the final threshold; Through the above steps, the final threshold can be determined, so as to more accurately reflect the quality level of the current batch of propolis liquid and improve the accuracy and reliability of the detection.

[0027] At this time, the content of the result output and warning becomes: During the actual detection process, compare the signature components under any number in the propolis liquid of the batch to be detected with their corresponding final thresholds; if a single type of signature component exceeds the standard, a first-level warning signal is issued; if multiple types of signature components (more than 1 type of signature component) exceed the standard, a second-level warning signal is issued; Among them, the first-level warning signal can be given by the flashing of a yellow signal light; The second-level warning signal can be given by the flashing of an orange signal light; If a single type of signature component exceeds the standard, a text message needs to be sent to notify the staff; If multiple types of signature components exceed the standard, the machine needs to be stopped for inspection.

[0028] Specifically, through the dynamic cognitive digital twin technology, the threshold can be corrected through simulation feedback. By introducing a correction coefficient and comparing the feedback rates, it is selectively determined whether correction is needed according to the simulation feedback results. On the one hand, it can avoid the reduction of detection efficiency caused by redundant correction actions. On the other hand, when it is found that the correction feedback is better, the adaptive threshold can be corrected for the second time, which can more comprehensively reflect the impact of parameter adjustment on the adaptive threshold, thereby further improving the accuracy and reliability of the detection; supporting real-time remote monitoring and warning functions, it can timely detect quality fluctuations and issue warning signals. This function not only improves the flexibility of quality control but also significantly enhances the practicality of quality management; Generally speaking, the dynamic cognitive digital twin technology, with its unique feedback correction mechanism and real-time monitoring ability, optimizes the detection efficiency while ensuring the detection accuracy.

[0029] Example 3: This embodiment provides a data processing system for detecting the quality of propolis liquid, which includes: Data acquisition and processing module: Obtain the spectral data of the propolis liquid to be detected in a batch, analyze the content of signature components, and obtain production stage parameters; preprocess the spectral data; Feature engineering construction module: Extract spectral features based on the preprocessed spectral data and construct a multi-dimensional vector matrix; Multi-model fusion analysis module: According to the multi-dimensional vector matrix, use the SVM classifier to judge compliance; according to the extracted spectral features, use the isolation forest to judge anomalies; summarize the judgment results, and based on the pre-built rule engine, obtain the pass rate of the current batch. Compare with the pass rate of the previous batch to analyze and obtain the quality change trend; Dynamic threshold decision module: Construct an adaptive threshold model. On the premise of determining the initial threshold, adjust the initial threshold according to the quality change trend result to obtain the adaptive threshold corresponding to different types of signature components; Digital twin construction module: Run the constructed digital twin of the production process and analyze the production stage parameters to determine whether to trigger the correction mechanism. Introduce a correction coefficient into the virtual model in the digital twin for dynamic compensation; Compare the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, and determine the final threshold according to the comparison result; Result output and warning module: During the actual detection process, conduct comparative analysis based on the final threshold and decide whether to issue a warning signal.

[0030] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0031] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0032] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for processing quality inspection data of propolis liquid, and the steps of the method are as follows: Obtain spectral data of the batch of propolis liquid to be tested, analyze the content of characteristic components, and obtain production stage parameters; Preprocess the spectral data; Extract spectral features based on the preprocessed spectral data and construct a multi-dimensional vector matrix; it is characterized in that it further includes: According to the multi-dimensional vector matrix, use the SVM classifier to judge compliance; according to the extracted spectral features, use the Isolation Forest to judge anomalies; summarize the judgment results, and based on the pre-built rule engine, obtain the pass rate of the current batch, and analyze the quality change trend compared with the pass rate of the previous batch; Construct an adaptive threshold model, and on the premise of determining the initial threshold, make a decision to adjust the initial threshold according to the quality change trend result to obtain the adaptive threshold corresponding to different types of signature components; Run the constructed digital twin of the production process, and analyze the production stage parameters to judge whether to trigger the correction mechanism, and introduce a correction coefficient into the virtual model in the digital twin for dynamic compensation; compare the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, and determine the final threshold in turn according to the comparison results; In the actual detection process, conduct a comparative analysis according to the final threshold and decide whether to issue a warning signal.

2. A method for processing propolis liquid quality detection data according to claim 1, characterized in that: The types of signature components at least include: flavonoids and terpenoids; the production stage parameters at least include: the adjustment direction and magnitude of the temperature, pH value, and feeding speed of the propolis liquid to be detected in the production stage.

3. A method for processing propolis liquid quality detection data according to claim 2, characterized in that: Preprocessing the spectral data at least includes: spectral noise reduction and baseline correction; Spectral feature extraction: Use the characteristic peak localization algorithm to detect the characteristic peaks corresponding to flavonoids and terpenoids and calculate the peak height ratio; the number of dimensions in the multi-dimensional vector matrix is the same as the number of types of signature components.

4. A method for processing propolis liquid quality detection data according to claim 1, characterized in that: After summarizing the judgment results, the process of running the rule engine is as follows: Screen the propolis liquid samples that are both qualified and normal in category label from the propolis liquid to be detected, and count the proportion of the number of screened samples in the total number of the propolis liquid to be detected to obtain the pass rate; When comparing with the pass rate of the previous batch of propolis liquid: If the pass rate of the previous batch of propolis liquid exceeds the pass rate of the propolis liquid to be detected, it is determined that the quality change trend is a downward trend; otherwise, the quality change trend is an upward trend.

5. A method for processing propolis liquid quality detection data according to claim 1, characterized in that: The initial threshold is μ ± s×σ; where μ and σ respectively represent the mean and standard deviation corresponding to the same type of signature component; the value range of s is: s > 0.

6. A method for processing propolis liquid quality detection data according to claim 4, characterized in that: When the quality change trend is an upward trend, the formula based on which the dynamic adjustment model is: ； Meaning of letters: T_new: Adaptive threshold, w1: Weight 1, T_old: Initial threshold, w2: Weight 2, and 1 > w1 > w2 > 0, μ_current: Mean of the batch to be detected; When the quality change trend is a downward trend, the formula based on which the dynamic adjustment model is: ； ； Meaning of letters: T_new_max: Upper limit value of the adaptive threshold, T_old_max: Upper limit value of the initial threshold, k: Expansion factor, and k > 0, T_new_min: Lower limit value of the adaptive threshold, T_old_min: Lower limit value of the initial threshold, σ_current: Mean of the batch to be detected.

7. A method for processing propolis liquid quality detection data according to claim 6, characterized in that: Under the condition of adjusting the parameters of any production stage, the correction mechanism is triggered; In the correction mechanism, the correction coefficient introduced by dynamic compensation is obtained by running a pre-built correction function model: Collect the mean μ1 and standard deviation σ1 of the contents of different types of signature components in the previous several batches of propolis liquid, and calculate the difference Δμ between the mean μ1 and μ_current, and the difference Δσ between the standard deviation σ1 and σ_current for the same type of signature component; Among them, both Δμ and Δσ are greater than 0; Obtain the corresponding correction coefficient Q for the batch to be detected, and perform the Sigmoid function correction action based on Δμ and Δσ; The corrected adaptive threshold is Q×T_new.

8. A method for processing propolis liquid quality detection data according to claim 7, characterized in that: In the comparison of the corrected adaptive threshold and the feedback rate simulated under the original adaptive threshold, the comparison process is as follows: When the feedback rates corresponding to the corrected adaptive threshold all exceed the feedback rate corresponding to the original adaptive threshold, the original adaptive threshold is maintained and used as the final threshold; When the feedback rate corresponding to any corrected adaptive threshold does not exceed the feedback rate corresponding to the original adaptive threshold, the corresponding corrected adaptive threshold is called and used as the final threshold; When the feedback rates corresponding to the corrected adaptive thresholds do not exceed the feedback rate corresponding to the original adaptive threshold, compare the feedback rates corresponding to the two corrected adaptive thresholds, and select the one with the smaller feedback rate as the final threshold; Among them, the feedback rate is either the false alarm rate or the missed alarm rate.

9. A method for processing propolis liquid quality detection data according to claim 1, characterized in that: The process of comparative analysis based on the final threshold is as follows: Compare the signature component under any number in the propolis liquid of the batch to be detected with its corresponding final threshold: When a single type of signature component exceeds the standard, a first-level warning signal is issued; When multiple types of signature components exceed the standard, a second-level warning signal is issued.

10. A data processing system for quality inspection of propolis liquid, the system includes: Data acquisition and processing module: Obtain the spectral data of the propolis liquid of the batch to be detected, analyze the content of signature components, and obtain production stage parameters; Preprocess the spectral data; Feature engineering construction module: Extract spectral features based on the preprocessed spectral data and construct a multi-dimensional vector matrix; characterized in that it further includes: Multi-model fusion analysis module: According to the multi-dimensional vector matrix, use the SVM classifier to judge compliance; according to the extracted spectral features, use the isolation forest to judge anomalies; summarize the judgment results, and based on a pre-built rule engine, obtain the pass rate of the current batch, and analyze the quality change trend compared with the pass rate of the previous batch; Dynamic threshold decision module: Construct an adaptive threshold model, and on the premise of determining the initial threshold, make a decision to adjust the initial threshold according to the quality change trend result to obtain the adaptive thresholds corresponding to different types of signature components; Digital twin construction module: Run the constructed digital twin of the production process, analyze the production stage parameters to judge whether to trigger the correction mechanism, introduce a correction coefficient in the virtual model in the digital twin for dynamic compensation; compare the corrected adaptive threshold with the feedback rate simulated under the original adaptive threshold, and determine the final threshold according to the comparison result; Result output warning module: During the actual detection process, it conducts comparative analysis based on the final threshold and decides whether to issue a warning signal.

Images