Cosmetics Antiseptic Ingredient Monitoring System

Abstract

The present invention relates to the technical field of ingredient monitoring, specifically a monitoring system for anti-corrosion ingredients in cosmetics. The system includes an ingredient data acquisition module, an ingredient anomaly detection module, an ingredient pedigree analysis module, and a monitoring and feedback adjustment module. In the present invention, by introducing the sliding window technique and the locally weighted scatterplot smoothing model, the real-time performance and data accuracy of anti-corrosion ingredient monitoring are enhanced, the real-time tracking ability of changes in preservative concentration is improved, allowing the system to quickly respond to minor changes. The isolation forest algorithm is used to analyze mutation points in time series data, providing an efficient anomaly detection method for cosmetics. This algorithm is suitable for big data environments, significantly improving the processing speed and accuracy. Combining near-infrared spectroscopy technology and Kalman filtering technology, the accuracy of predicting preservative concentration is improved, enabling manufacturers to accurately control the amount of preservatives in cosmetics, meet safety standards, and reduce the risk of overuse.

Description

Technical Field

[0001] The present invention relates to the technical field of ingredient monitoring, and in particular to a cosmetic preservative ingredient monitoring system. Background Art

[0002] Ingredient monitoring technology involves the use of various sensors and analytical equipment to detect and determine the composition of substances. This technology is widely used in many industries such as food safety, environmental monitoring, medical diagnosis, and cosmetics manufacturing. In the cosmetics industry, ingredient monitoring technology ensures the quality and safety of products by accurately measuring the content of active ingredients in the products, including preservatives, antioxidants and other additives. This technology usually includes the use of spectroscopy, chromatography, mass spectrometry and electrochemical analysis to perform quantitative and qualitative analysis to ensure that all ingredients meet strict safety standards.

[0003] Among them, the cosmetic preservative ingredient monitoring system is a system specially designed to track and analyze the preservative content in cosmetics. Preservatives are an important component in cosmetics and are used to prevent the growth of microorganisms and extend the shelf life of products. Its main purpose is to ensure that the amount of preservatives used in cosmetics is within safety standards while avoiding exceeding the maximum limit allowed by regulations. By real-time monitoring of the chemical composition of products, it can help manufacturers maintain product quality, protect consumer safety, and comply with industry regulations.

[0004] When monitoring ingredients such as preservatives that change rapidly and are sensitive to small changes in concentration, traditional systems are unable to capture key data in a timely manner, which can easily lead to missed detections or false detections. In addition, traditional technologies are cumbersome to operate and have long analysis cycles, making them unsuitable for fast-paced production environments, affecting production efficiency and the ability to respond to market changes in a timely manner. For example, for products with new formulas or rapidly changing ingredients, traditional systems find it difficult to provide efficient quality assurance, which can easily affect the product's market response speed and consumer trust. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose a cosmetic preservative ingredient monitoring system.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a cosmetic preservative ingredient monitoring system, the system comprising:

[0007] The ingredient data collection module uses sliding window technology to calculate the statistical characteristics of the data in each window based on the continuously collected preservative ingredient concentration data, and establishes a local weighted scatter point smoothing model based on the reaction characteristics of the preservative types, including phenoxyethanol, ethylhexylglycerin, phenylpropanol and caprylyl glycol, to generate trend model data;

[0008] Based on the trend model data, the component anomaly detection module uses the isolation forest algorithm to analyze the mutation points in the time series, identify the abnormal anti-corrosion component concentration changes in the data, and for the identified abnormal data points, compare the historical data patterns, refine the anomaly types, and generate a component anomaly feature set;

[0009] Based on near-infrared spectroscopy technology, the component pedigree analysis module collects the spectral data of cosmetic samples, performs multivariate analysis, compares the spectral analysis results with the machine vision data, and through Kalman filtering, optimizes the estimated amount of anti-corrosion component concentration to generate a component concentration optimization index;

[0010] Based on the component anomaly feature set and the component concentration optimization index, the monitoring and feedback adjustment module constructs a dynamic update mechanism to adjust the monitoring threshold, updates the monitoring parameter settings in real time, optimizes the monitoring and early warning process of the preservative type and component stability, and generates a component monitoring feedback result.

[0011] The improvement of the present invention is that the calculation steps of the statistical characteristics of the data within the window are specifically as follows:

[0012] Based on the continuously collected anti-corrosion component concentration data, set the sliding window size, collect the anti-corrosion component concentration data sequence, slice the data sequence according to the window size, and use the formula:

[0013]

[0014] Calculate the required number of windows n to generate data subsets, where S is the collected anti-corrosion component concentration data sequence and w is the sliding window size;

[0015] For each of the data subsets, calculate the statistic using the formula:

[0016]

[0017] and

[0018]

[0019] Obtain the average value μ of the i-th window i and the standard deviation σ i , where x is a data point in D i , D i is the data subset, p i is the weight of the data point x, determined according to the time order of the data points, λ is the smoothing coefficient used to adjust the sensitivity of the standard deviation, and w is the sliding window size;

[0020] Combine the average value and the standard deviation into a statistical characteristic vector of the data subset, and the formula is:

[0021] Vi =(μ i ,σ i )

[0022] Obtain the data statistical characteristics V within the sliding window i , where μ i is the mean value, σ i is the standard deviation, and iterative analysis is performed based on V i , including outlier detection or trend prediction.

[0023] The improvement of the present invention is that the steps of establishing the locally weighted scatter smoothing model are specifically as follows:

[0024] Based on the data statistical characteristics, select the smoothing parameter and determine the bandwidth of the local weight, and adjust the weight function. The formula is:

[0025] w(x, h, β)=exp(-β·(x 2 / h 2 )

[0026] Generate the adjusted weight function w(x, h, β), where h represents the bandwidth, which determines the local range of weight distribution, β represents the attenuation coefficient, which adjusts the sensitivity of the weight function, and x is the independent variable distance of the data point;

[0027] Based on the adjusted weight function, apply the LOWESS formula:

[0028]

[0029] Among them, Calculate the locally weighted regression value of each data point x i to obtain the locally weighted scatter smoothing model, represents the predicted value obtained by LOWESS calculation, w ij represents the local weight of point x j to point x i , n represents the total number of data points, x i and x j respectively represent the independent variable values of the i-th and j-th data points, y j represents the dependent variable value of the j-th data point, |x i -x j | is the distance on the independent variable axis between point i and point j, and w is the weight function.

[0030] The improvement of the present invention is that the steps of identifying the change in the concentration of the anti-corrosion component are specifically as follows:

[0031] Based on the data of the trend model, use the isolation forest algorithm to analyze the data points in the time series, and adopt the formula:

[0032]

[0033] Calculate the anomaly score for each data point to obtain a list of preliminarily identified anomaly points. Among them, S(x) is the anomaly score of data point x, d(x) is the distance from the data point to the adjacent decision boundary, m is the average distance from the data point to the decision boundary, κ is the density adjustment coefficient, p(x) is the prior probability of data point x, and σ is the standard deviation of the data point;

[0034] Based on the list of anomaly points, set a threshold to filter data points with high anomaly scores, using the formula:

[0035] E(x) = {x | S(x) > θ × α}

[0036] Determine the anomaly data points exceeding the threshold, and generate a list of data points with high anomaly scores. Among them, θ is the threshold for identifying high anomaly scores, α is the threshold adjustment factor, E(x) is the set of anomaly data points, S(x) is the anomaly score of data point x, and x is the data point;

[0037] Verify the list of data points with high anomaly scores, check the context and adjacent values of the data points, using the formula:

[0038]

[0039] Generate the change characteristics of the anomaly anti-corrosion component concentration. Among them, V(x) is the set of verified anomaly data points, E is the set of anomaly data points, μ is the average value of the data points, used to determine the degree of anomaly, λ is the adjustment parameter, used to affect the anomaly sensitivity, and x is the data point.

[0040] The improvement of the present invention is that the comparison step of the historical data pattern is specifically as follows:

[0041] Collect the historical data of the change in the anti-corrosion component concentration. For each currently identified anomaly data point, calculate the difference degree from the historical data, using the formula:

[0042]

[0043] Generate a list of deviations D(x) between the current data point x and the historical pattern, where h i is the anti-corrosion component concentration value of the historical data point, N represents the total number of historical data points, w i is the weight of the i-th data point, γ represents the power exponent used to adjust the sensitivity of the deviation calculation, δ is the parameter used for smoothing, and x is the anti-corrosion component concentration value of the data point;

[0044] According to the list of deviations, analyze the types of deviations, including periodic changes, sudden increases or sudden decreases, and conduct anomaly type determination to obtain the component anomaly feature set.

[0045] The improvement of the present invention is that the specific steps for collecting the spectral data are as follows:

[0046] Based on the near-infrared spectroscopy technology, set the parameters of the near-infrared spectrometer, adjust the light source intensity and detector sensitivity, match the characteristics of the cosmetic sample, and use the formula:

[0047]

[0048] Generate the spectral device configuration parameter P, where L represents the light source intensity, η represents the device efficiency, S represents the sensitivity of the detector, A represents the absorbance of the sample, and c represents the optical path length;

[0049] According to the spectral device configuration parameter, collect the spectral data of the cosmetic sample, record the reflectivity of each sample, and use the formula:

[0050]

[0051] Calculate the reflectivity R of the sample to obtain the spectral data set of the cosmetic sample, where I represents the intensity of the reflected light, I0 represents the intensity of the incident light, and β represents the correction factor used to adjust the error in the reflectivity calculation;

[0052] Format the spectral data set into a data file, and use the data sorting formula:

[0053]

[0054] Generate the spectral data file D, where R n represents the reflectivity of the data point, γ is the normalization factor, and N represents the total number of data points.

[0055] The improvement of the present invention is that the specific steps for obtaining the component concentration optimization index are as follows:

[0056] Analyze the spectral data, identify the spectral characteristics of multiple components in the sample, and use the formula:

[0057]

[0058] Generate the multivariate analysis result M, where x i represents an observed value of a variable in the sample, w i represents the weight assigned to the i-th variable, α represents the power exponent used to amplify the influence of the observed value, and λ and δ i represent the attenuation coefficients, and N represents the total number of variables;

[0059] Compare the multivariate analysis result with the data obtained by machine vision, and use the formula:

[0060]

[0061] Generate differential analysis log C, where M j represents the result from multivariate analysis, V j represents the data from machine vision, K represents the total number of data points for comparison, and β and γ represent the exponents used to adjust the sensitivity of difference calculation;

[0062] Apply the Kalman filter to optimize the concentration of the anti-corrosion component predicted based on spectral and visual data, using the formula:

[0063] F k = F k-1 + K k ·(Z k - F k-1 ) g

[0064] Generate an optimized index for the component concentration, where F k represents the estimated concentration value, K k represents the Kalman gain, Z k represents the current observed value, F k-1 represents the previous estimated value, and g represents the adjustment exponent for the update step size.

[0065] The improvement of the present invention is that the step of obtaining the component monitoring feedback result is specifically:

[0066] Based on the component anomaly feature set and the optimized index for the component concentration, collect real-time monitoring data from the monitoring process adjusted by the dynamic update mechanism, using the formula:

[0067]

[0068] Generate real-time monitoring data, where D represents the total sum of the collected real-time monitoring data, c i represents the concentration of the anti-corrosion component at the i-th monitoring point, λ i represents the weight coefficient assigned according to the characteristics of the i-th monitoring point, and N represents the total number of monitoring points;

[0069] Evaluate the deviation of the real-time monitoring data from the threshold, using the formula:

[0070]

[0071] Generate monitoring data deviation evaluation information, where E represents the weighted average deviation between the monitoring data and the threshold, c i represents the concentration of the anti-corrosion component in real-time monitoring, T set represents the anti-corrosion component concentration threshold, w i represents the weight coefficient of the deviation, and N represents the total number of monitoring points;

[0072] According to the deviation evaluation information, use the formula:

[0073] F = "Adjust" if E > θ else "Maintain"

[0074] Generate the component monitoring feedback result F, where E is the weighted average deviation between the monitoring data and the threshold, and θ is the decision threshold.

[0075] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0076] In the present invention, by introducing the sliding window technology and the locally weighted scatterplot smoothing model, the real-time performance and data accuracy of the anti-corrosion component monitoring are enhanced, the real-time tracking ability of the change in the preservative concentration is improved, the system is allowed to quickly respond to minor changes, the isolation forest algorithm is used to analyze the mutation points in the time series data, providing an efficient anomaly detection method for cosmetics. This algorithm is suitable for the big data environment, significantly improving the processing speed and accuracy. By combining the near-infrared spectroscopy technology and the Kalman filtering technology, the accuracy of predicting the preservative concentration is improved, enabling the manufacturer to accurately control the amount of preservatives in cosmetics, meeting the safety standards and reducing the risk of overuse. Description of the Drawings

[0077] Figure 1 It is a module diagram of the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0078] Figure 2 It is a calculation flowchart of the data statistical characteristics within the window in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0079] Figure 3 It is a flowchart of establishing the locally weighted scatterplot smoothing model in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0080] Figure 4 It is an identification flowchart of the change in the anti-corrosion component concentration in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0081] Figure 5 It is a comparison flowchart of the historical data patterns in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0082] Figure 6 It is a flowchart of collecting spectral data in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0083] Figure 7 It is a flowchart of obtaining the optimization index of the component concentration in the cosmetics anti-corrosion component monitoring system proposed by the present invention;

[0084] Figure 8This is the flowchart for obtaining the monitoring feedback results of the components in the cosmetic anti-corrosion component monitoring system proposed by the present invention. Detailed implementation mode

[0085] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0086] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0087] Embodiment

[0088] Please refer to Figure 1 , the present invention provides a technical solution: the cosmetic anti-corrosion component monitoring system includes:

[0089] Based on the continuously collected anti-corrosion component concentration data, the component data acquisition module uses the sliding window technology to calculate the data statistical characteristics within each window, and establishes a locally weighted scatter plot smoothing model for the reaction characteristics of the preservative types, including phenoxyethanol, ethylhexylglycerin, phenylpropanol and octyldodecanol, to generate trend model data; the locally weighted scatter plot smoothing model is used to capture the non-linear influence of the preservative types on the concentration change, so as to provide a more accurate basis for trend analysis and anomaly warning.

[0090] Based on the trend model data, the component anomaly detection module uses the isolation forest algorithm to analyze the mutation points in the time series, identify the abnormal anti-corrosion component concentration changes in the data, and compare the historical data patterns for the identified abnormal data points to refine the anomaly types and generate a component anomaly feature set;

[0091] Based on the near-infrared spectroscopy technology, the component pedigree analysis module collects the spectral data of the cosmetic samples, performs multivariate analysis, compares the spectral analysis results with the machine vision data, and optimizes the estimated amount of the anti-corrosion component concentration through Kalman filtering to generate a component concentration optimization index; the machine vision data is collected by the camera equipment on the production line and is used to monitor the color and texture of the cosmetics, etc.

[0092] The monitoring and feedback adjustment module constructs a dynamic update mechanism to adjust the monitoring threshold based on the component anomaly feature set and the component concentration optimization index, updates the monitoring parameter settings in real time, optimizes the monitoring and warning process of the preservative type and component stability, and generates the component monitoring feedback result.

[0093] The trend model data includes the concentration change trend, seasonal adjustment result, and anomaly warning index. The component anomaly feature set includes the identified anomaly events, anomaly degree evaluation results, and influencing factor analysis results. The component concentration optimization index includes the concentration stability score information, prediction accuracy information, and adjustment response sensitivity.

[0094] Please refer to Figure 2 , and the calculation steps of the statistical characteristics of the data within the window are specifically as follows:

[0095] Based on the continuously collected preservative component concentration data, set the sliding window size, collect the preservative component concentration data sequence, slice the data sequence according to the window size, and use the formula:

[0096]

[0097] Calculate the required number of windows n to generate data subsets, where S is the collected preservative component concentration data sequence and w is the sliding window size;

[0098] For each data subset, calculate the statistic using the formula:

[0099]

[0100] and

[0101]

[0102] Obtain the average value μ of the i-th window i and the standard deviation σ i , where x is a data point in D i , D i is the data subset, p i is the weight of the data point x, determined according to the time order of the data points, λ is the smoothing coefficient used to adjust the sensitivity of the standard deviation, and w is the sliding window size;

[0103] Combine the average value and the standard deviation into a statistical characteristic vector of the data subset, and the formula is:

[0104] V i =(μ i ,σ i )

[0105] Obtain the statistical characteristics V of the data within the sliding window i , where μ iis the average value, and σ i is the standard deviation, and iterative analysis is performed according to V i including outlier detection or trend prediction.

[0106] In calculating the number of windows, assume the following data:

[0107] Assume that the window size w = 10 (one window contains 10 data points), and there is a data sequence S with a length |S| = 150.

[0108] Calculation process:

[0109] Determine the window size w = 10;

[0110] Measure the data sequence length |S| = 150;

[0111] Apply the formula to calculate

[0112] which means dividing the data sequence into 15 windows.

[0113] In calculating the weighted average and adjusted standard deviation, assume the following data:

[0114] Assume the weight p i is assigned according to the position of the data point in the window (for example, earlier data points have lower weights), and λ (smoothing coefficient) is set to 2, and the example values of the data points x in the window are [5, 10, 15, 20, 25, 30, 35, 40, 45, 50].

[0115] Calculation process:

[0116] For each data point x in the window, set the value of p i For example, p i = 0.1i (where i is the index position of the data point in the window);

[0117] Calculate the weighted average:

[0118]

[0119] Calculate the adjusted standard deviation:

[0120] First, calculate the cube of the difference between each data point and the average value:

[0121] For example, |5 - 33.5| 3 and |10 - 33.5| 3 and so on.

[0122] Apply the formula to calculate the standard deviation:

[0123]

[0124] Weighted average μ i = 33.5 represents the average value of the data points within the window considering the weights;

[0125] Adjusted standard deviation σ i = 29.8 represents the degree of dispersion of the data point distribution, and after adjustment, it reflects greater sensitivity;

[0126] Apply formula V i = (μ i , σ i ), combines the average value and the standard deviation into a statistical feature vector, representing the statistical features of the data within the window.

[0127] Please refer to Figure 3 , and the steps to establish the locally weighted scatter plot smoothing model are specifically as follows:

[0128] Based on the data statistical features, select the smoothing parameter and determine the bandwidth of the local weights, and adjust the weight function. The formula is:

[0129] w(x, h, β) = exp(-β · (x 2 / h 2 ))

[0130] Generate the adjusted weight function w(x, h, β), where h represents the bandwidth, determining the local range of weight distribution, β represents the attenuation coefficient, adjusting the sensitivity of the weight function, and x is the independent variable distance of the data point;

[0131] Based on the adjusted weight function, apply the LOWESS formula:

[0132]

[0133] where, Calculate the locally weighted regression value of each data point x i , and obtain the locally weighted scatter plot smoothing model. represents the predicted value obtained through LOWESS calculation, w ij represents the local weight of point x j for point x i , n represents the total number of data points, x i and x j represent the independent variable values of the i-th and j-th data points respectively, y j represents the dependent variable value of the j-th data point, |x i - x j | is the distance on the independent variable axis between point i and point j, and w is the weight function.

[0134] Suppose there is the following data:

[0135] Assume h = 5 (bandwidth, affecting the local smoothing range) and β = 0.5 (attenuation coefficient, adjusting the sensitivity of the weight function) are selected, and the weight at x = 3 will be calculated.

[0136] Calculation process:

[0137] Determine the bandwidth h = 5 and the attenuation coefficient β = 0.5;

[0138] Substitute x = 3 into the weight function to calculate the weight:

[0139] w(3, 5, 0.5) = exp(-0.5·(3 2 / 5 2 ))

[0140] = exp(-0.18) = 0.835

[0141] It means that the weight of the point x j at a distance of 3 units from the prediction point x i is 0.835, indicating that the nearer points have a greater influence on the predicted value.

[0142] Suppose there is the following data:

[0143] Assume n = 3, and the values of the data point positions x and the dependent variable y are as follows:

[0144] x1 = 0, y1 = 2

[0145] x2 = 3, y2 = 3

[0146] x3 = 5, y3 = 5

[0147] Predict the value at x i = 3

[0148] Calculation process:

[0149] Calculate the weight w for each data point ij , for example, for j = 1:

[0150]

[0151] Calculate using the weight and the corresponding y value

[0152]

[0153] Finally, the value at the prediction point x i = 3 is approximately 5.183, and this value represents the value of the dependent variable y predicted according to the LOWESS model at x = 3.

[0154] Please refer toFigure 4 , the steps for identifying the change in the concentration of the anti-corrosion component are specifically as follows:

[0155] Based on the trend model data, using the Isolation Forest algorithm, analyze the data points in the time series, and adopt the formula:

[0156]

[0157] Calculate the anomaly score of each data point to obtain a list of initially identified anomaly points. Among them, S(x) is the anomaly score of data point x, d(x) is the distance from the data point to the adjacent decision boundary, m is the average distance from all data points to the decision boundary, κ is the density adjustment coefficient considering the crowding degree around the data point, p(x) is the prior probability of data point x, and σ is the standard deviation of the data point;

[0158] Based on the list of anomaly points, set a threshold to screen the data points with high anomaly scores, and use the formula:

[0159] E(x) = {x | S(x) > … × α}

[0160] Determine the abnormal data points exceeding the threshold to generate a list of data points with high anomaly scores. Among them, θ is the threshold used to identify high anomaly scores, α is the threshold adjustment factor to adapt to different data characteristics, E(x) is the set of abnormal data points, S(x) is the anomaly score of data point x, and x is the data point;

[0161] Verify the list of data points with high anomaly scores, check the context and adjacent values of the data points, and adopt the formula:

[0162]

[0163] Generate the characteristics of abnormal anti-corrosion component concentration changes. Among them, V(x) is the set of verified abnormal data points, E is the set of abnormal data points, μ is the average value of the data points used to determine the degree of abnormality, λ is the adjustment parameter used to affect the anomaly sensitivity, and x is the data point.

[0164] Suppose there is the following data:

[0165] The position of data point x is x = 10;

[0166] The distance to the adjacent decision boundary d(x) = 2;

[0167] The average distance m = 5;

[0168] The density coefficient κ of the data point = 0.5;

[0169] The prior probability p(x) of the data point = 0.1;

[0170] The standard deviation σ = 1.5.

[0171] Calculation process:

[0172] Calculate d(x)+κ·p(x)=2 + 0.5×0.1 = 2.05;

[0173] Calculate m+σ = 5 + 1.5 = 6.5;

[0174] Substitute the value into the anomaly score formula:

[0175]

[0176] It shows that the anomaly score of the data point x = 10 is approximately 0.729, and this value can be used to compare with the anomaly threshold in the subsequent steps to determine whether it is an anomaly.

[0177] Suppose there is the following data:

[0178] Anomaly score threshold θ = 0.6;

[0179] Adjustment factor α = 1.2.

[0180] Calculation process:

[0181] Calculate the adjusted threshold θ×α = 0.6×1.2 = 0.72;

[0182] Compare the above anomaly score S(x)=0.729 with the adjusted threshold 0.72;

[0183] Since 0.729>0.72, the data point x = 10 is confirmed as an anomaly;

[0184] It shows that after adjusting the threshold, the data point x = 10 is confirmed as an anomaly.

[0185] Suppose there is the following data:

[0186] The average value μ of the data point x is 8;

[0187] Adjustment parameter λ = 2.

[0188] Calculation process:

[0189] For the data point x = 10, calculate |x - μ| = |10 - 8| = 2;

[0190] Calculate λ·μ = 2×8 = 16;

[0191] Substitute into the verification formula:

[0192]

[0193] The calculation shows that the verified anomaly score is 0.875, indicating a relatively high anomaly of the data point x = 10.

[0194] Please refer to Figure 5 , the specific steps for comparing historical data patterns are as follows:

[0195] Collect historical data on the change in the concentration of the anti-corrosion component. For each currently identified abnormal data point, calculate the difference degree from the historical data using the formula:

[0196]

[0197] Generate a list D(x) of the deviations of the current data point x from the historical pattern, where h i is the concentration value of the anti-corrosion component at the historical data point, N represents the total number of historical data points, w i is the weight of the i-th data point, and based on time proximity, the closer to the present, the greater the weight. γ represents the power exponent used to adjust the sensitivity of the deviation calculation, and δ is the parameter used for smoothing to ensure calculation stability. x is the concentration value of the anti-corrosion component at the data point;

[0198] According to the deviation list, analyze the types of deviations, including periodic changes, sudden increases or decreases, and perform abnormal type determination to obtain the component abnormal feature set.

[0199] Suppose there is the following data:

[0200] Suppose there is a current data point x = 50 and a small historical data set that contains three points h =

[0201] 45, 55, 60];

[0202] The weights of the points are w = [0.5, 1, 0.5], and the selected γ = 2, and the smoothing parameter δ = 1.

[0203] Calculation process:

[0204] Setting of weights and historical points:

[0205] h1 = 45, w1 = 0.5

[0206] h2 = 55, w2 = 1

[0207] h3 = 60, w3 = 0.5

[0208] Calculate the weighted deviation of each historical point:

[0209] For h1:

[0210] w1·|x - h1| γ = 0.5·|50 - 45| 2 = 0.5·25 = 12.5

[0211] For h2:

[0212] w2·|x - h2| γ = 1·|50 - 55| 2 = 1·25 = 25

[0213] For h3:

[0214] w3·|x - h3| γ = 0.5·|50 - 60| 2 = 0.5·100 = 50

[0215] Calculation of the sum and average deviation:

[0216] Total deviation:

[0217] 12.5 + 25 + 50 = 87.5

[0218] Denominator:

[0219] N + δ = 3 + 1 = 4

[0220] Calculate D(x):

[0221]

[0222] The calculation result D(x) = 21.875 represents the average weighted deviation of the current data point x = 50 relative to the given historical data pattern. A high value indicates a significant difference between the current data point and the historical data, indicating abnormal activity or change.

[0223] Please refer to Figure 6 , the specific steps for collecting spectral data are as follows:

[0224] Based on near-infrared spectroscopy technology, set the parameters of the near-infrared spectrometer, adjust the light source intensity and detector sensitivity to match the characteristics of the cosmetic sample, and use the formula:

[0225]

[0226] Generate the spectral device configuration parameter P, where L represents the light source intensity, η represents the device efficiency, which measures how the device performance converts the light source energy, S represents the detector sensitivity, A represents the absorbance of the sample, and c represents the optical path length, which affects the distance that light travels in the device;

[0227] According to the spectral device configuration parameter, collect the spectral data of the cosmetic sample, record the reflectance of each sample, and use the formula:

[0228]

[0229] Calculate the reflectance R of the sample to obtain the spectral data set of the cosmetic sample, where I represents the reflected light intensity, I0 represents the incident light intensity, and β represents the correction factor, which is used to adjust the error in the reflectance calculation;

[0230] Format the spectral data set into a data file using the data arrangement formula:

[0231]

[0232] Generate the spectral data file D, where R n represents the reflectance of the data point, γ is the normalization factor to ensure data consistency and comparability, and N represents the total number of data points.

[0233] Suppose there is the following data:

[0234] Suppose the light source intensity L = 100 units;

[0235] The device efficiency η = 0.8 (80% efficiency);

[0236] The detector sensitivity S = 0.9;

[0237] The absorbance A of the sample = 1.2;

[0238] The optical path length c = 2 meters.

[0239] Calculation process

[0240] Device parameter calculation:

[0241] Calculate L·η·S = 100×0.8×0.9 = 72;

[0242] Calculate A·c = 1.2×2 = 2.4;

[0243] Substitute the values into the device configuration parameter formula:

[0244]

[0245] The configuration parameter P of the spectral device is 30, which is a synthetic parameter reflecting the overall sensitivity and efficiency of the device.

[0246] Suppose there is the following data:

[0247] The intensity of the reflected light I = 50 units;

[0248] The intensity of the incident light I0 = 100 units;

[0249] The correction factor β = 1.1.

[0250] Reflectance calculation:

[0251] Calculate the corrected incident light intensity:

[0252] I0·β = 100×1.1 = 110

[0253] Calculate the reflectance:

[0254]

[0255] The reflectance R of the sample is 0.455, indicating that a certain proportion of light is reflected by the sample.

[0256] Suppose there is the following data:

[0257] Reflectance data points R = [0.45, 0.50, 0.48];

[0258] Normalization factor γ = 0.5;

[0259] Total number of data points N = 3.

[0260] Data arrangement:

[0261] Normalize each reflectance and sum them up:

[0262]

[0263] It is shown that the sum of the sorted spectral data file D is 2.86, which can be used for further analysis or storage.

[0264] Please refer to Figure 7 , and the specific steps for obtaining the composition concentration optimization index are as follows:

[0265] Analyze the spectral data, identify the spectral characteristics of multiple components in the sample, and use the formula:

[0266]

[0267] Generate the multivariate analysis result M, where x i represents an observed value of a variable in the sample, w i represents the weight assigned to the i-th variable, α represents the power exponent used to amplify the influence of the observed value, and λ and δ i represent the attenuation coefficients used to simulate the reduction of the influence of environmental factors or time on the variable, and N represents the total number of variables;

[0268] Compare the multivariate analysis result with the data obtained by machine vision, and use the formula:

[0269]

[0270] Generate the difference analysis log C, where M j represents the result from the multivariate analysis, V j represents the data from machine vision, K represents the total number of data points for comparison, and β and γ represent the exponents used to adjust the sensitivity of the difference calculation;

[0271] Apply Kalman filtering to optimize the concentration of anti-corrosion components predicted based on spectral and visual data, using the formula:

[0272] F k = F k-1 + K k ·(Z k - F k-1 ) g

[0273] Generate an optimized index for the component concentration, where F k represents the estimated concentration value, K k represents the Kalman gain, which is used to adjust the degree to which the estimated value approaches the actual observed value, Z k represents the current observed value, F k-1 represents the previous estimated value, and g represents the adjustment exponent for the update step.

[0274] Suppose there is the following data:

[0275] x i Suppose it is the spectral intensity, and the sample values are [100, 200, 150];

[0276] Suppose the weight w i is [0.5, 1.0, 0.75];

[0277] α is set to 2;

[0278] λ is set to 0.05;

[0279] δ i is set to [1, 2, 3];

[0280] N3.

[0281] Calculation process:

[0282] For each i:

[0283] Calculate

[0284] [100 2 ,200 2 ,150 2 = [10000, 40000, 22500]

[0285] Calculate

[0286] [e -0.05·1 ,e -0.05·2 ,e -0.05·3 = [0.951, 0.905, 0.860]

[0287] Calculate each item

[0288] [0.5 · 10000 · 0.951, 1.0 · 40000 · 0.905, 0.75 · 22500 · 0.860]

[0289] = [4755, 36200, 14512.5]

[0290] Calculate the sum M:

[0291] 4755 + 36200 + 14512.5 = 55467.5

[0292] The result of the multivariate analysis M = 55449.5 represents the comprehensive analysis value obtained from the weighted and adjusted spectral intensities. A higher value indicates a high concentration or strong response of a specific component.

[0293] Suppose there is the following data:

[0294] M j Suppose it is [55450, 55460, 55455];

[0295] V j Suppose it is [55430, 55470, 55440];

[0296] K is 3;

[0297] β is set to 1.5;

[0298] γ is set to 0.5.

[0299] Calculation process:

[0300] For each j:

[0301] Calculate the absolute value of the difference and weight it:

[0302] |M j - V j | β = |20| 1.5 , |10| 1.5 , |15| 1.5 ≈ [89, 32, 58]

[0303] Calculate the sum and normalize it:

[0304]

[0305] The differential analysis report C = 103.35 represents the average adjusted difference value between the multivariate analysis result and the machine vision data, which is used to evaluate the consistency or difference between the two technologies.

[0306] Please refer to Figure 8, the steps for obtaining the component monitoring feedback results are specifically as follows:

[0307] Based on the component anomaly feature set and the component concentration optimization index, collect real-time monitoring data from the monitoring process adjusted by the dynamic update mechanism, and use the formula:

[0308]

[0309] Generate real-time monitoring data, where D represents the total sum of the collected real-time monitoring data, c i represents the anti-corrosion component concentration at the i-th monitoring point, and λ i represents the weight coefficient assigned according to the characteristics of the i-th monitoring point, and N represents the total number of monitoring points;

[0310] Evaluate the deviation of the real-time monitoring data from the threshold, and use the formula:

[0311]

[0312] Generate monitoring data deviation evaluation information, where E represents the weighted average deviation between the monitoring data and the threshold, c i represents the anti-corrosion component concentration of the real-time monitoring, T set represents the anti-corrosion component concentration threshold, and w i represents the weight coefficient of the deviation, which is used to adjust the influence of data at different monitoring points, and N represents the total number of monitoring points;

[0313] According to the deviation evaluation information, use the formula:

[0314] F = "Adjust" if E > θ else "Maintain"

[0315] Generate the component monitoring feedback result F, where E is the weighted average deviation between the monitoring data and the threshold, and θ is the decision threshold, which is used to determine whether it is necessary to adjust the monitoring strategy.

[0316] Suppose there is the following data:

[0317] c i is the anti-corrosion component concentration at the i-th monitoring point. Suppose the concentrations at the three monitoring points are 15 ppm, 20 ppm, and 25 ppm respectively;

[0318] λ i The set weights are 0.5, 1.0, and 1.5, reflecting the importance of different monitoring points;

[0319] N is set to 3.

[0320] For each monitoring point:

[0321] c1·λ1 = 15 × 0.5 = 7.5

[0322] c2·λ2 = 20×1.0 = 20

[0323] c3·λ3 = 25×1.5 = 37.5

[0324] Calculate the sum:

[0325] D = 7.5 + 20 + 37.5 = 65

[0326] The sum D = 65 represents the sum of the weighted real-time monitoring data, which is the total concentration weighted based on the importance of the monitoring points.

[0327] Suppose there is the following data:

[0328] c i are 15 ppm, 20 ppm, and 25 ppm respectively;

[0329] T set is supposed to be 18 ppm;

[0330] w i are 0.5, 1.0, and 1.5;

[0331] N is 3.

[0332] |c1 - T set |·w1 = |15 - 18|×0.5 = 1.5

[0333] |c2 - T set |·w2 = |20 - 18|×1.0 = 2.0

[0334] |c3 - T set |·w3 = |25 - 18|×1.5 = 10.5

[0335] Calculate the weighted average deviation:

[0336]

[0337] The weighted average deviation E = 4.67 represents the average deviation of the monitoring data relative to the preset threshold, taking into account the importance of different monitoring points.

[0338] The weighted average deviation obtained for E is 4.67, and θ is supposed to be 3.

[0339] Check if E is greater than θ:

[0340] Compare E = 4.67 with θ = 3. Since 4.67 > 3, F is for adjustment.

[0341] It indicates that the monitoring strategy needs to be adjusted because the deviation shown by the real-time monitoring data exceeds the set decision threshold.

[0342] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A monitoring system for cosmetic preservatives, characterized in that, The system includes: Based on the continuously collected data of the concentration of anti-corrosion components, the component data acquisition module uses the sliding window technology to calculate the statistical characteristics of the data within each window, and establishes a locally weighted scatterplot smoothing model for the reaction characteristics of the types of preservatives, including phenoxyethanol, ethylhexylglycerin, phenylpropanol, and octylglycol, to generate trend model data; The specific steps for establishing the locally weighted scatterplot smoothing model are as follows: Based on the statistical characteristics of the data, select the smoothing parameter and the bandwidth for determining the local weight, and adjust the weight function. The formula is: ； Generate the adjusted weight function , where represents the bandwidth, which determines the local range of weight allocation, represents the attenuation coefficient, which adjusts the sensitivity of the weight function, is the independent variable distance of the data point; Based on the adjusted weight function, apply the LOWESS formula: ； Among them, calculate the locally weighted regression value of each data point to obtain a locally weighted scatter plot smoothing model, represents the predicted value obtained by LOWESS calculation, represents the point for the point the local weight of, represents the total number of data points, and respectively represent the and the independent variable values of the th data point, represents the dependent variable value of the th data point, is the distance on the independent variable axis between the point and the point , is the weight function; Based on the trend model data, the component anomaly detection module uses the isolation forest algorithm to analyze the mutation points in the time series, identify the abnormal changes in the concentration of anti-corrosion components in the data, and for the identified abnormal data points, compare the historical data patterns, refine the types of anomalies, and generate a component anomaly feature set; Based on the near-infrared spectroscopy technology, the component pedigree analysis module collects the spectral data of the cosmetic samples, performs multivariate analysis, compares the spectral analysis results with the machine vision data, and optimizes the estimated amount of the anti-corrosion component concentration through Kalman filtering to generate a component concentration optimization index; The specific steps for obtaining the component concentration optimization index are as follows: Analyze the spectral data to identify the spectral characteristics of multiple components in the sample. Use the formula: ； Generate multivariate analysis results , where represents an observed value of a variable in the sample, represents the weight assigned to the th variable, represents the power exponent used to amplify the influence of the observed value, and represent the attenuation coefficient, represents the total number of variables; Compare the multivariate analysis results with the data obtained by machine vision. Use the formula: ； Generate differential analysis log , where represents the result from multivariate analysis, represents the data from machine vision, represents the total number of data points for comparison, and represents the index used to adjust the sensitivity of difference calculation; Apply the Kalman filtering to optimize the concentration of the anti-corrosion component predicted based on the spectral and visual data. Use the formula: ； Generate the optimization index of component concentration, where represents the estimated concentration value, represents the Kalman gain, represents the current observed value, represents the previous estimated value, represents the adjustment index of the update step size; Based on the component anomaly feature set and the component concentration optimization index, the monitoring and feedback adjustment module constructs a dynamic update mechanism to adjust the monitoring threshold, updates the monitoring parameter settings in real time, optimizes the monitoring and warning process of the types of preservatives and the component stability, and generates a component monitoring feedback result.

2. The cosmetic preservative ingredient monitoring system according to claim 1, characterized in that The specific steps for calculating the statistical characteristics of the data within the window are as follows: Based on the continuously collected data of the concentration of anti-corrosion components, set the size of the sliding window, collect the data sequence of the concentration of anti-corrosion components, slice the data sequence according to the window size. Use the formula: ； Calculate the number of windows required , generate data subsets, where is the data sequence of the collected anti-corrosion component concentration is the sliding window size; For each data subset, calculate the statistic. Use the formula: and ; Obtain the average value of the th window and the standard deviation where is a data point in which is a data subset is the weight of the data point determined according to the chronological order of the data points is a smoothing coefficient used to adjust the sensitivity of the standard deviation is the sliding window size; Combine the average value and the standard deviation into a statistical characteristic vector of the data subset. The formula is: ； Obtain the statistical characteristics of the data within the sliding window , where is the average value, is the standard deviation, and iterative analysis is performed according to including outlier detection or trend prediction.

3. The cosmetic preservative component monitoring system according to claim 1, wherein The specific steps for identifying the change in the concentration of the anti-corrosion component are as follows: Based on the trend model data, use the isolation forest algorithm to analyze the data points in the time series. Use the formula: ； Calculate the anomaly score for each data point to obtain a list of preliminarily identified anomaly points, where, is the anomaly score of the data point . is the distance from the data point to the neighboring decision boundary, is the average distance from the data point to the decision boundary, is the density adjustment coefficient, is the prior probability of the data point , is the standard deviation of the data point; Based on the list of anomaly points, set a threshold to screen the data points with high anomaly scores. Use the formula: ； Determine abnormal data points exceeding the threshold, and generate a list of data points with high anomaly scores, where is the threshold for identifying high anomaly scores, is the threshold adjustment factor, is the set of abnormal data points, is the data point of the anomaly score, is the data point; Verify the list of data points with high anomaly scores, and check the context and adjacent values of the data points. Use the formula: ； Generate the concentration change characteristics of abnormal anti-corrosion components, where is a set of verified abnormal data points, is a set of abnormal data points, is the average value of data points, used to determine the degree of abnormality, is an adjustment parameter, used to affect the abnormality sensitivity, is a data point.

4. The cosmetic preservative ingredient monitoring system according to claim 1, wherein The specific steps for comparing the historical data patterns are as follows: Collect the historical data of the change in the concentration of the anti-corrosion component. For each currently identified abnormal data point, calculate the difference degree from the historical data. Use the formula: ； Generate the current data point List of deviations from the historical pattern , where is the concentration value of the anti-corrosion component of the historical data point, represents the total number of historical data points, is the weight of the th data point, represents the power exponent used to adjust the sensitivity of the deviation calculation, is the parameter for smoothing, is the concentration value of the anti-corrosion component of the data point; According to the list of deviations, analyze the types of deviations, including periodic changes, sudden increases or sudden decreases, and perform anomaly type determination to obtain a component anomaly feature set.

5. The cosmetic anti-corrosion component monitoring system according to claim 1, wherein The specific steps for collecting the spectral data are as follows: Based on the near-infrared spectroscopy technology, set the parameters of the near-infrared spectrometer, adjust the light source intensity and detector sensitivity, match the characteristics of the cosmetic samples, and use the formula: ； Generate spectral device configuration parameters , where represents the light source intensity, represents the device efficiency, represents the sensitivity of the detector, represents the absorbance of the sample, represents the optical path length; According to the configuration parameters of the spectral device, collect the spectral data of the cosmetic samples, record the reflectivity of each sample, and use the formula: ； Calculate the reflectance of the sample , to obtain the spectral data set of the cosmetic sample, where represents the reflected light intensity represents the incident light intensity represents the correction factor, which is used to adjust the error in the reflectance calculation; Format the spectral data set into a data file and use the data sorting formula: ； Generate a spectral data file , where represents the reflectance of the data point, is the normalization factor, represents the total number of data points.

6. The cosmetic preservative component monitoring system according to claim 1, wherein, The specific steps for obtaining the component monitoring feedback result are as follows: Based on the component anomaly feature set and the component concentration optimization index, collect real-time monitoring data from the monitoring process adjusted by the dynamic update mechanism, and use the formula: ； Generate real-time monitoring data, including: Represents the total amount of real-time monitoring data collected. Representative The concentration of antiseptic components at each monitoring point, Representatives according to The weight coefficient of the characteristic allocation of each monitoring point, represents the total number of monitoring points; Evaluate the deviation of the real-time monitoring data from the threshold and use the formula: ； Generate monitoring data deviation evaluation information, where represents the weighted average deviation between the monitoring data and the threshold value, represents the concentration of the anti-corrosion component in real-time monitoring, represents the threshold value of the anti-corrosion component concentration, represents the weight coefficient of the deviation, represents the total number of monitoring points; According to the deviation evaluation information, use the formula: ； Generate component monitoring feedback results , where is the weighted average deviation between the monitoring data and the threshold value, is the decision threshold.

Images