Intelligent metering calibration method and system

Abstract

The invention relates to the technical field of calibration tests, in particular to an intelligent metering calibration method and system, and the method comprises the following steps: calculating the similarity between data points, constructing a similarity matrix, further calculating the fitting degree, carrying out the spectral decomposition, extracting a feature vector, carrying out the dimension reduction, mapping to a low-dimensional space, carrying out the data point clustering based on a feature vector matrix, and carrying out the calibration of the data points. An error cluster is identified and an error pattern is determined. According to the method, by calculating the similarity between the data points and mapping the similarity to the low-dimensional space, the error modes can be efficiently recognized and classified, the error correction path can be dynamically adjusted, error correction is more flexible and real-time, the error analysis depth is improved through spectral decomposition and feature vector extraction, and the error analysis accuracy is improved. Therefore, the error trends of the equipment under different conditions are accurately identified, the measurement precision and stability are remarkably improved, the error correction path parameters are updated in real time, the adaptability of the system to the operation change of the equipment is enhanced, and the calibration time is effectively shortened.

Description

Technical Field

[0001] The present invention relates to the field of calibration test technology, and in particular to an intelligent metrology calibration method and system. Background Art

[0002] The field of calibration test technology is a technical research direction for testing and calibrating various types of equipment, instruments or systems. It mainly involves the measurement and verification of mechanical properties, electrical characteristics, thermodynamic parameters or other physical quantities to ensure the accuracy and stability of measurement equipment and systems. This technology is widely used in manufacturing, energy, aerospace, medical equipment, environmental monitoring and other industries to improve equipment performance, optimize operating parameters, and ensure safety and quality through precise calibration tests.

[0003] Among them, intelligent metrology calibration method refers to the technical method of optimizing the calibration process of metrology equipment and improving calibration efficiency and accuracy by using automation, information or intelligent means. Its main uses include completing high-precision calibration tasks in complex measurement environments, supporting multi-parameter synchronous measurement, and improving the work efficiency of laboratory or on-site calibration. It is widely used in high-precision instrument manufacturing, testing laboratories and metrology management of smart factories.

[0004] The existing technology mainly relies on traditional error correction methods for data processing, lacking in-depth analysis of error patterns and real-time dynamic adjustment. The existing solutions cannot effectively identify the variable error types in complex environments during the calibration process, resulting in the inability to correct errors in a timely manner, affecting the stability of equipment performance. In the long-term use of high-precision equipment, traditional methods are difficult to adapt to fluctuations in equipment performance, increasing the calibration cycle and the risk of error accumulation, affecting measurement accuracy and system reliability. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings in the prior art and to propose an intelligent measurement calibration method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an intelligent measurement calibration method, comprising the following steps: S1: Based on the instrument output data, calculate the Euclidean distance or cosine similarity between each group of data points, obtain the similarity value between each pair of data points, quantify the similarity between data points, and construct a similarity matrix; S2: Based on the similarity matrix, calculate the fit of each data point with other data points, construct a Laplace matrix, perform spectral decomposition, extract eigenvectors and reduce the dimension, obtain the distribution of data points in the low-dimensional space, and generate an eigenvector matrix; S3: After mapping the data points to a low-dimensional space based on the eigenvector matrix, clustering the data points according to similarity, identifying error clusters, assigning similar data points to the same error cluster, identifying the error pattern corresponding to each cluster, and generating error pattern clustering results; S4: based on the error pattern clustering result, select local data in each error cluster, fit the regression function, calculate the regression coefficient and adjust the error value, calculate the error correction path, and generate an error correction model; S5: Based on the error correction model, according to real-time feedback, extract real-time error data, calculate the error change trend, update the correction path parameters, readjust the weight coefficient of the correction path, dynamically adjust the correction path, and generate the error correction model update result.

[0007] The similarity matrix includes the Euclidean distance, cosine similarity, and similarity quantization value between each pair of data points; the eigenvector matrix includes the distribution of low-dimensional space, eigenvectors, and spectral decomposition results; the error pattern clustering results include error clusters, error patterns, and error cluster allocations; the error correction model includes regression function, regression coefficient, error correction path, and correction path weight; the error correction model update results include correction path, weight coefficient, error change trend, and correction path parameter update.

[0008] As a further solution of the present invention, the step of obtaining the similarity matrix is ​​specifically as follows: S111: Based on the output data of the instrument, the Euclidean distance between each group of data points is calculated, the multi-dimensional values ​​of each pair of data points are selected, the differences are calculated and squared, the squares of the differences are accumulated and the results are squared to generate the Euclidean distance value of each pair of data points; S112: Calculate the dot product and modular product of each pair of data points according to the Euclidean distance value of each pair of data points in combination with the multi-dimensional value vector of the data points, and perform an operation using the ratio of the two to obtain a cosine similarity value between each pair of data points; S113: Combining the Euclidean distance value of each pair of data points with the cosine similarity value between each pair of data points, a weighted combination is performed using the formula: ; Calculate the comprehensive similarity value of each pair of data points and construct a similarity matrix; in, Represents the similarity matrix Line The value of the column, Indicates Data points and The Euclidean distance between data points is Indicates Data points and The cosine similarity value between data points is is the cosine similarity weight adjustment coefficient, which is used to control the influence of cosine similarity. is the distance weight adjustment coefficient, which is used to balance the impact of distance attenuation on similarity calculation. As the distance nonlinear adjustment term, the absolute value is introduced to improve the robustness.

[0009] As a further solution of the present invention, the step of obtaining the eigenvector matrix is ​​specifically as follows: S211: Based on the similarity matrix, analyze the relationship between each data point and other data points, calculate the degree of fit, extract the degree information of the data point from the row data of the matrix, obtain the total degree by accumulating each row, and establish a Laplace matrix according to the degree value; S212: performing spectral decomposition from the Laplace matrix by a numerical analysis method, extracting key eigenvalues ​​and corresponding eigenvectors, selecting the dimension of dimensionality reduction as k, and combining the eigenvectors of the first k smallest non-zero eigenvalues ​​to obtain a set of dimensionality reduction feature vectors; S213: operating the dimension-reduced feature vector set using the formula: ; Calculate the position of each pair of data points in the current low-dimensional space, generate the eigenvector matrix, and complete the data dimensionality reduction process; in, Indicates the first eigenvector in the current eigenvector matrix Line The value of the column, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Representative The weight parameter of the feature vector is used to adjust the influence of the feature vector. Represents the exponential decay adjustment coefficient, which is used to control the decay amplitude of the distance between data points. Represents the normalized adjustment coefficient, which is used to normalize the result of the eigenvector matrix. Represents an exponential decay function based on Euclidean distance, which is used to measure the distance relationship between data points in the feature space. Represents the eigenvector matrix and How related the data points are along all dimensions.

[0010] As a further solution of the present invention, the step of obtaining the error pattern clustering result is specifically as follows: S311: Mapping the data points to a low-dimensional space based on the eigenvector matrix, using a distance metric to calculate the distance between the data points in the low-dimensional space, and using a clustering algorithm to preliminarily group the data points to generate a preliminary clustering result; S312: Analyze the preliminary clustering results, identify and mark data points with high similarity as error clusters, use statistical analysis methods to determine the data distribution characteristics of each cluster, and obtain error cluster marking results; S313: According to the error cluster marking result, a pattern recognition algorithm is applied to each error cluster, using the formula: ; Calculate the pattern characteristics of each error cluster, comprehensively refer to the weights of the data points in the cluster, and generate the error pattern clustering results; in, Indicates The pattern characteristics of the error clusters, Represents the cluster The feature vector of the data point, For data points In cluster The weight in reflects its contribution to the cluster mode. is a smoothing parameter to avoid the denominator being zero, The error cluster The total number of data points in .

[0011] As a further solution of the present invention, the step of obtaining the error correction model is specifically as follows: S411: selecting a set of local data points of each error cluster from the error pattern clustering result, applying linear regression analysis to the data to fit a local regression function, calculating an initial regression coefficient, and generating a preliminary regression model; S412: Analyze the preliminary regression model, adjust the error value to reduce the prediction deviation according to the regression analysis result, calculate the current regression coefficient, re-evaluate the error value and optimize the adjustment coefficient, and obtain the adjusted regression model; S413: Based on the adjusted regression model, mathematical optimization technology is applied to iteratively optimize the error correction path, using the formula: ; Generate error correction paths and build error correction models; in, represents the error correction model, is the regression coefficient for each data point, representing the weight of the point in the model, is the actual observed value, is the predicted value, Represents the model adjustment coefficient, which is used to linearly correct the deviation between the predicted value and the actual value. is the nonlinear adjustment index, represents the error correction term calculated according to the new model, Provides non-linear adjustment capability.

[0012] As a further solution of the present invention, the step of obtaining the error correction model update result is specifically: S511: extracting real-time error data based on the error correction model, calculating the change range of the error point in the current state, analyzing the correlation between the data, and generating error change trend analysis results; S512: calling the error change trend analysis result, dynamically updating the parameters in the correction path, adjusting the weight coefficient of the path, recalculating the correction path in combination with the current error distribution, and generating a correction path update result; S513: In combination with the corrected path update result, a nonlinear optimization formula is applied to the dynamically adjusted weight coefficient, using the formula: ; Calculate the dynamically adjusted weights of the correction path, and combine them with the dynamically updated weights to generate the error correction model update results; in, represents the modified path weight after dynamic adjustment, is the initial weight coefficient for each error point, is the real-time error change value, represents the smoothing parameter that limits the amplitude of nonlinear adjustments, represents the smoothing parameter that limits the range of dynamic influence, is the nonlinear gain coefficient of the error change, which is used to enhance the effect of the change amplitude on the weight adjustment.

[0013] An intelligent metrology calibration system, the intelligent metrology calibration system is used to perform the above-mentioned intelligent metrology calibration method, the system comprises: The data processing module calculates the Euclidean distance or cosine similarity between each group of data points based on the instrument output data, obtains the similarity value between each pair of data points, quantifies the similarity between data points, and constructs a similarity matrix; The feature extraction module calculates the fit between each data point and other data points based on the similarity matrix, constructs a Laplace matrix, performs spectral decomposition, extracts eigenvectors, performs dimensionality reduction, obtains the distribution of data points in low-dimensional space, and generates a eigenvector matrix; The error pattern recognition module maps the data points to a low-dimensional space based on the eigenvector matrix, clusters the data points according to similarity, identifies error clusters, assigns similar data points to the same error cluster, identifies the error pattern corresponding to each cluster, and generates error pattern clustering results; The error correction module selects local data in each error cluster based on the error pattern clustering result, fits the regression function, calculates the regression coefficient and adjusts the error value, calculates the error correction path, generates the error correction model, and extracts real-time error data according to real-time feedback, calculates the error change trend, updates the correction path parameters, readjusts the weight coefficient of the correction path, dynamically adjusts the correction path, and generates the error correction model update result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by calculating the similarity between data points and mapping them to a low-dimensional space, error patterns can be efficiently identified and classified. This method can dynamically adjust the error correction path, making error correction more flexible and real-time. Spectral decomposition and eigenvector extraction enhance the depth of error analysis, thereby accurately identifying the error trend of the device under different conditions, significantly improving measurement accuracy and stability. Real-time updating of error correction path parameters enhances the system's adaptability to changes in device operation and effectively shortens calibration time. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 is a flow chart of the steps of obtaining the similarity matrix of the present invention; Figure 3 Flow chart of the steps of obtaining the characteristic vector matrix of the present invention; Figure 4 A flowchart of the steps for obtaining the error pattern clustering results of the present invention; Figure 5 is a flow chart of the steps of obtaining the error correction model of the present invention; Figure 6 This is a flow chart of the steps for obtaining the update results of the error correction model of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is 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 intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are 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 cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Example 1: Please refer to Figure 1 The present invention provides a technical solution: an intelligent measurement calibration method, comprising the following steps: S1: Based on the instrument output data, calculate the Euclidean distance or cosine similarity between each group of data points, obtain the similarity value between each pair of data points, quantify the similarity between data points, and construct a similarity matrix; S2: Based on the similarity matrix, calculate the fit of each data point with other data points, construct the Laplace matrix, and perform spectral decomposition to extract eigenvectors and reduce the dimension, obtain the distribution of data points in the low-dimensional space, and generate the eigenvector matrix; S3: After mapping the data points to the low-dimensional space based on the eigenvector matrix, cluster the data points according to similarity, identify error clusters, assign similar data points to the same error cluster, identify the error pattern corresponding to each cluster, and generate error pattern clustering results; S4: Based on the error pattern clustering results, select local data in each error cluster, fit the regression function, calculate the regression coefficient and adjust the error value, calculate the error correction path, and generate the error correction model; S5: Based on the error correction model, according to the real-time feedback, extract the real-time error data, calculate the error change trend, update the correction path parameters, readjust the weight coefficient of the correction path, dynamically adjust the correction path, and generate the error correction model update result.

[0019] The similarity matrix includes the Euclidean distance, cosine similarity, and similarity quantization value between each pair of data points. The eigenvector matrix includes the distribution of low-dimensional space, eigenvectors, and spectral decomposition results. The error pattern clustering results include error clusters, error patterns, and error cluster allocations. The error correction model includes regression function, regression coefficient, error correction path, and correction path weight. The error correction model update results include correction path, weight coefficient, error change trend, and correction path parameter update.

[0020] See also Figure 2 , the steps to obtain the similarity matrix are as follows: S111: Based on the output data of the instrument, the Euclidean distance between each group of data points is calculated, the multi-dimensional values ​​of each pair of data points are selected, the differences are calculated and squared, the squares of the differences are accumulated and the results are squared to generate the Euclidean distance value of each pair of data points; Based on the instrument output data, calculating the Euclidean distance between each group of data points is a basic but critical step. First, extract the coordinates between each pair of data points, calculate the coordinate differences and then square them. Then sum the squared results and finally take the square root of the sum to get the Euclidean distance between the two points. This distance value is the basis for subsequent similarity calculations. The accurate calculation of the Euclidean distance ensures the accuracy of the cosine similarity calculation. Through this method, the straight-line distance between any two points can be accurately measured in multidimensional space, which is crucial for subsequent data analysis and pattern recognition. Through accurate distance calculations, we can better understand the relative position and similarity between data points.

[0021] S112: Calculate the dot product and modular product of each pair of data points according to the Euclidean distance value of each pair of data points in combination with the multi-dimensional value vector of the data points, and perform an operation using the ratio of the two to obtain the cosine similarity value between each pair of data points; According to the Euclidean distance between data points, the cosine similarity of the data points is further analyzed. The specific process includes performing dot product operations on each pair of data point vectors and calculating the modular product of the two vectors. The ratio of the dot product and the modular product is the cosine similarity. The calculation of cosine similarity involves the comparison of vector directions, not just sizes. This makes cosine similarity particularly suitable for measuring the similarity between text or other forms of data points. Through cosine similarity, data points that are similar in direction but different in magnitude can be effectively identified, thereby providing support for complex data analysis.

[0022] S113: Combining the Euclidean distance value of each pair of data points with the cosine similarity value between each pair of data points, a weighted combination is performed using the formula: ; Calculate the comprehensive similarity value of each pair of data points and construct a similarity matrix; in, Represents the similarity matrix Line The value of the column, Indicates Data points and The Euclidean distance between data points is Indicates Data points and The cosine similarity value between data points is is the cosine similarity weight adjustment coefficient, which is used to control the influence of cosine similarity. is the distance weight adjustment coefficient, which is used to balance the impact of distance attenuation on similarity calculation. As the distance nonlinear adjustment term, the absolute value is introduced to improve the robustness.

[0023] formula: ; The benefit of the formula is that, by combining the square adjustment of the Euclidean distance and the weighted cosine similarity, the balance of the model's sensitivity to different types of data is enhanced, especially in high-dimensional data, this balance can help better capture the intrinsic structure of the data.

[0024] Detailed explanation of the formula and the process of formula calculation and derivation: Assume two data points and The coordinates in two-dimensional space are and ,calculate ,set up , ,set up and , put it into the formula and calculate: ; The results show that after considering the comprehensive factors of distance and angle, the two data points and The similarity is 3.2286, which provides a quantitative similarity score for this pair of points when constructing the similarity matrix, facilitating further data analysis and decision support.

[0025] See also Figure 3 , the specific steps for obtaining the eigenvector matrix are: S211: Based on the similarity matrix, analyze the relationship between each data point and other data points, calculate the degree of fit, extract the degree information of the data point from the row data of the matrix, obtain the total degree by accumulating each row, and establish a Laplace matrix according to the degree value; The process involves analyzing the row data of the similarity matrix to extract the connection strength of each data point, then accumulating these values ​​to form the degree information of each point, and then using this degree information to create a Laplacian matrix. The key to this step is how to accurately quantify the connection strength between data points and effectively convert these quantified values ​​into degrees. This requires not only proper processing and analysis of the data, but also ensuring that all numerical values ​​accurately reflect the actual relationship between data points.

[0026] S212: performing spectral decomposition from the Laplace matrix by numerical analysis method, extracting key eigenvalues ​​and corresponding eigenvectors, selecting the dimension of dimensionality reduction as k, combining the eigenvectors of the first k smallest non-zero eigenvalues, and obtaining a set of dimensionality reduction eigenvectors; Key eigenvectors are extracted from the Laplace matrix through spectral decomposition for data dimensionality reduction and further analysis. In this process, the choice of spectral decomposition method is crucial because it directly affects the accuracy of eigenvalues ​​and eigenvectors, as well as the effect of dimensionality reduction. The key lies in how to select appropriate eigenvectors and combine them correctly to form the data expression after dimensionality reduction. Through spectral decomposition, the eigenvectors corresponding to the first k smallest non-zero eigenvalues ​​are selected and combined to form a set of reduced dimensionality feature vectors for subsequent analysis. These vector sets mathematically represent the most important structural features of the data, laying the foundation for further processing and analysis of the data.

[0027] S213: Operate on the reduced dimension feature vector set using the formula: ; Calculate the position of each pair of data points in the current low-dimensional space, generate the eigenvector matrix, and complete the data dimensionality reduction process; in, Indicates the first eigenvector in the current eigenvector matrix Line The value of the column, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Representative The weight parameter of the feature vector is used to adjust the influence of the feature vector. Represents the exponential decay adjustment coefficient, which is used to control the decay amplitude of the distance between data points. Represents the normalized adjustment coefficient, which is used to normalize the result of the eigenvector matrix. Represents an exponential decay function based on Euclidean distance, which is used to measure the distance relationship between data points in the feature space. Represents the eigenvector matrix and How related the data points are along all dimensions.

[0028] formula: ; The benefit of the formula is that, by introducing exponential decay and normalization, the model's sensitivity to differences between data points is enhanced while maintaining the overall stability of the data structure, effectively reflecting the subtle differences between different data points, and reducing the impact of noise while maintaining data diversity.

[0029] Detailed explanation of the formula and the process of formula calculation and derivation: set up , , , , , calculate each term: ; ; ; Substituting into the formula: ; This result shows that with Increase, The value will tend to a stable ratio, which reflects that after data dimensionality reduction, the relative position relationship between different data points is clearer, which is helpful for further data analysis and pattern recognition.

[0030] See also Figure 4 , the specific steps for obtaining the error pattern clustering results are: S311: Mapping the data points to a low-dimensional space based on the eigenvector matrix, using a distance metric to calculate the distance between the data points in the low-dimensional space, and using a clustering algorithm to preliminarily group the data points to generate a preliminary clustering result; In low-dimensional space, the distance between data points is calculated using a distance measurement formula. This calculation is based on the eigenvector after dimensionality reduction. The Euclidean distance of each pair of data points in the low-dimensional feature space is calculated to determine their mutual similarity or difference. The distance matrix obtained is the input of the clustering algorithm. Clustering algorithms such as k-means or DBSCAN are used to perform preliminary grouping of data points based on similarity. In the clustering process, the distance from each point to each cluster center is calculated, and the nearest cluster center is selected as the label of the point, thereby assigning the data points to the corresponding groups and generating preliminary clustering results.

[0031] S312: Analyze the preliminary clustering results, identify and mark data points with high similarity as error clusters, use statistical analysis methods to determine the data distribution characteristics of each cluster, and obtain error cluster marking results; By analyzing the preliminary clustering results, data points with high similarity can be identified. These data points are marked as error clusters. The identification of error clusters depends on the evaluation of clustering quality. For example, the silhouette coefficient is used to evaluate the compactness and separation of clusters. Clustering results with high compactness and low separation indicate the existence of error clusters. Statistical analysis is performed on each cluster to determine the distribution characteristics of the data points within the cluster, which involves calculating the mean and variance of the data points within the cluster. The data distribution pattern of each cluster is marked according to these statistical characteristics to generate a marking result.

[0032] S313: According to the error cluster labeling result, a pattern recognition algorithm is applied to each error cluster, using the formula: ; Calculate the pattern characteristics of each error cluster, comprehensively refer to the weights of the data points in the cluster, and generate the error pattern clustering results; in, Indicates The pattern characteristics of the error clusters, Represents the cluster The feature vector of the data point, For data points In cluster The weight in reflects its contribution to the cluster mode. is a smoothing parameter to avoid the denominator being zero, The error cluster The total number of data points in .

[0033] formula: ; The benefit of the formula is that it calculates the pattern characteristics of the error cluster by weighted averaging, which can accurately reflect the common characteristics of the data points in the cluster and reduce the impact of a single abnormal data point.

[0034] Detailed explanation of the formula and the process of formula calculation and derivation: Suppose there is an error cluster containing 5 data points, and their eigenvectors are , weight They are , smoothing parameter The eigenvector value of each data point is obtained through dimensionality reduction in the previous step, and the weight is assigned according to the density and position of the data point in the dimensionality reduction space. The calculation process is as follows: ; ; ; ; This result shows that the weighted average after weight adjustment can accurately represent the common characteristics of data points in the error cluster, thereby effectively guiding subsequent data processing and analysis decisions.

[0035] See also Figure 5 , the specific steps of obtaining the error correction model are: S411: Selecting a set of local data points of each error cluster from the error pattern clustering result, applying linear regression analysis to the data to fit a local regression function, calculating an initial regression coefficient, and generating a preliminary regression model; When selecting a set of data points from the error pattern clustering results and performing linear regression analysis, we must first ensure that the selected data points are representative and cover the core characteristics of the error cluster. The selection of these data points needs to be based on the statistical analysis results of the error cluster, mainly considering the distribution density and outliers of the data points. After the data points are extracted, their features will be standardized to eliminate the influence of different scales, and then the regression coefficients of these data points will be calculated. This calculation not only includes the numerical calculation of the coefficients, but also the evaluation of their statistical significance to determine the effectiveness of the model. After the regression analysis is completed, the generated preliminary regression model will clarify the correlation of the data points in each error cluster and the fit of the regression line. This preliminary model provides a basis for subsequent error adjustment and model optimization.

[0036] S412: Analyze the preliminary regression model, adjust the error value to reduce the prediction deviation according to the regression analysis result, calculate the current regression coefficient, re-evaluate the error value and optimize the adjustment coefficient, and obtain the adjusted regression model; The preliminary regression model makes error adjustments based on the results of regression analysis. The key to this step is how to effectively use the results of regression analysis to adjust errors in order to reduce prediction bias. First, the prediction error of each data point is analyzed, and the data points where the error exceeds expectations are identified. These points are analyzed in depth to understand the cause of the error, whether it is a problem with data collection or an inaccurate model. Then, the parameters of the regression model are adjusted based on these analysis results. The adjustments include re-evaluating the factors that affect the regression coefficient, adjusting the coefficient value, and adding or deleting model variables. Through these adjustments, a more accurate regression model is obtained, which is closer to the actual data distribution, reduces prediction errors, and improves the prediction accuracy of the model.

[0037] S413: After synthesizing the adjusted regression model, mathematical optimization techniques are applied to iteratively optimize the error correction path using the formula: ; Generate error correction paths and build error correction models; in, represents the error correction model, is the regression coefficient for each data point, representing the weight of the point in the model, is the actual observed value, is the predicted value, Represents the model adjustment coefficient, which is used to linearly correct the deviation between the predicted value and the actual value. is the nonlinear adjustment index, represents the error correction term calculated according to the new model, Provides non-linear adjustment capability.

[0038] formula: ; The formula is beneficial because it introduces a nonlinear adjustment parameter and error scaling factor , improves the fitting ability of complex data distribution, and combines the data point weight Dynamically adjusting the contribution of each point to the revised model ensures the flexibility and adaptability of the model.

[0039] Detailed explanation of the formula and the process of formula calculation and derivation: 1. Determine the values ​​of each parameter: , obtained through weight analysis, , obtain the actual observation value through the data acquisition system, , the initial prediction value obtained by the prediction model, , the error scaling factor determined by model optimization, , nonlinear adjustment parameters obtained through data characteristic analysis.

[0040] 2. Enter the formula for calculation: calculate ; calculate ; calculate .

[0041] 3. Summarize the results: Assumptions ,but .

[0042] The result shows that the current error correction value is -13.03, and the corresponding correction path reflects that the deviation between the predicted value and the actual value needs further adjustment. This result provides a parameter basis for constructing an error correction model.

[0043] See also Figure 6 , the specific steps for obtaining the updated results of the error correction model are: S511: extracting real-time error data based on the error correction model, calculating the change range of the error point in the current state, analyzing the correlation between the data, and generating error change trend analysis results; Based on the error correction model, real-time error data is extracted. By calling the change amplitude value of the current state of the error point, the acquisition equipment is used to record the change trend of each error point, the correlation between the data is calculated and the change trend matrix is ​​constructed. In the process of trend analysis, the extracted data points are first normalized, the similarity between the error points is calculated, and the classification of the current trend direction is judged by the amplitude of the error change. The trend distribution model is generated to obtain the error change trend analysis results.

[0044] S512: calling the error change trend analysis result, dynamically updating the parameters in the correction path, adjusting the weight coefficient of the path, recalculating the correction path in combination with the current error distribution, and generating a correction path update result; According to the error change trend analysis results, the classification information in the trend distribution model is called to analyze the distribution parameters of each type of error, including the mean, variance and range of the error points. Based on these parameters, the parameter adjustment amount of the correction path is dynamically updated, and the adjusted path weight is calculated. After redistributing the weight, the value of the correction path is normalized. By updating the parameters and weights of the correction path, the dynamic update of the correction path is completed, and the correction path update result is generated.

[0045] S513: Combined with the updated result of the modified path, a nonlinear optimization formula is applied to the dynamically adjusted weight coefficient, using the formula: ; Calculate the dynamically adjusted weights of the correction path, and combine them with the dynamically updated weights to generate the error correction model update results; in, represents the modified path weight after dynamic adjustment, is the initial weight coefficient for each error point, is the real-time error change value, represents the smoothing parameter that limits the amplitude of nonlinear adjustments, represents the smoothing parameter that limits the range of dynamic influence, is the nonlinear gain coefficient of the error change, which is used to enhance the effect of the change amplitude on the weight adjustment.

[0046] formula: ; The benefit of the formula is that the introduction of gain coefficients and nonlinear adjustment items enhances the sensitivity of the correction path to dynamic changes in errors, while smoothing excessive adjustment amplitudes to ensure computational stability.

[0047] Detailed explanation of the formula and the process of formula calculation and derivation: set up , which contains 5 error points, and its weight coefficient , error change value They are , gain factor , smoothing parameter . Substituting these parameters into the formula: ; Step by step calculation: ; ; ; The result shows that the weight value of the dynamically adjusted correction path is 0.494, which is directly used to update the generated error correction model results.

[0048] An intelligent metrology calibration system, which is used to perform the above-mentioned intelligent metrology calibration method, and the system includes: The data processing module calculates the Euclidean distance or cosine similarity between each group of data points based on the instrument output data, obtains the similarity value between each pair of data points, quantifies the similarity between data points, and constructs a similarity matrix; The feature extraction module calculates the fit of each data point with other data points based on the similarity matrix, constructs the Laplace matrix, performs spectral decomposition, extracts eigenvectors, performs dimensionality reduction, obtains the distribution of data points in low-dimensional space, and generates a eigenvector matrix; The error pattern recognition module maps the data points to a low-dimensional space based on the eigenvector matrix, clusters the data points according to similarity, identifies error clusters, assigns similar data points to the same error cluster, identifies the error pattern corresponding to each cluster, and generates error pattern clustering results; The error correction module selects local data in each error cluster based on the error pattern clustering results, fits the regression function, calculates the regression coefficient and adjusts the error value, calculates the error correction path, generates the error correction model, and extracts real-time error data according to real-time feedback, calculates the error change trend, updates the correction path parameters, readjusts the weight coefficient of the correction path, dynamically adjusts the correction path, and generates the error correction model update results.

[0049] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent measurement calibration method, characterized in that: The following steps are involved: Based on the instrument output data, calculate the Euclidean distance or cosine similarity between each group of data points, obtain the similarity value between each pair of data points, quantify the similarity between data points, and construct a similarity matrix; Based on the similarity matrix, the degree of fit between each data point and other data points is calculated, a Laplace matrix is ​​constructed, and spectral decomposition is performed to extract eigenvectors and reduce the dimensionality, obtain the distribution of data points in the low-dimensional space, and generate an eigenvector matrix; Based on the eigenvector matrix, after mapping the data points to the low-dimensional space, clustering the data points according to similarity, identifying error clusters, assigning similar data points to the same error cluster, identifying the error pattern corresponding to each cluster, and generating error pattern clustering results; Based on the error pattern clustering result, local data in each error cluster is selected, a regression function is fitted, a regression coefficient is calculated and an error value is adjusted, an error correction path is calculated, and an error correction model is generated; Based on the error correction model, according to real-time feedback, real-time error data is extracted, the error change trend is calculated, the correction path parameters are updated, the weight coefficient of the correction path is readjusted, the correction path is dynamically adjusted, and the error correction model update result is generated.

2. The intelligent measurement and calibration method according to claim 1, characterized in that: The similarity matrix includes the Euclidean distance, cosine similarity, and similarity quantization value between each pair of data points; the eigenvector matrix includes the distribution of low-dimensional space, eigenvectors, and spectral decomposition results; the error pattern clustering results include error clusters, error patterns, and error cluster allocations; the error correction model includes regression function, regression coefficient, error correction path, and correction path weight; the error correction model update results include correction path, weight coefficient, error change trend, and correction path parameter update.

3. The intelligent measurement and calibration method according to claim 2, characterized in that: The steps of obtaining the similarity matrix are specifically as follows: Based on the output data of the instrument, the Euclidean distance between each group of data points is calculated, and the multi-dimensional values ​​of each pair of data points are selected to find the difference and square them. The squared differences are accumulated and the results are squared to generate the Euclidean distance value of each pair of data points. According to the Euclidean distance value of each pair of data points, combined with the multi-dimensional value vector of the data points, the dot product and the modular product of each pair of data points are calculated respectively, and the ratio of the two is used for operation to obtain the cosine similarity value between each pair of data points; Combine the Euclidean distance value of each pair of data points with the cosine similarity value between each pair of data points to perform a weighted combination, using the formula: ； Calculate the comprehensive similarity value of each pair of data points and construct a similarity matrix; in, Represents the similarity matrix Line The value of the column, Indicates Data points and The Euclidean distance between data points is Indicates Data points and The cosine similarity value between data points is is the cosine similarity weight adjustment coefficient, which is used to control the influence of cosine similarity. is the distance weight adjustment coefficient, which is used to balance the impact of distance attenuation on similarity calculation. As the distance nonlinear adjustment term, the absolute value is introduced to improve the robustness.

4. The intelligent measurement and calibration method according to claim 3, characterized in that: The steps for obtaining the eigenvector matrix are specifically as follows: Based on the similarity matrix, the relationship between each data point and other data points is analyzed, the degree of fit is calculated, the degree information of the data point is extracted from the row data of the matrix, the total degree is obtained by accumulating each row, and a Laplace matrix is ​​established according to the degree value; From the Laplace matrix, perform spectral decomposition by numerical analysis method, extract key eigenvalues ​​and corresponding eigenvectors, select the dimension of dimensionality reduction as k, combine the eigenvectors of the first k smallest non-zero eigenvalues, and obtain a set of dimensionality reduction eigenvectors; The reduced dimension feature vector set is operated using the formula: ； Calculate the position of each pair of data points in the current low-dimensional space, generate the eigenvector matrix, and complete the data dimensionality reduction process; in, Indicates the first eigenvector in the current eigenvector matrix Line The value of the column, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Represents the first eigenvector in the original eigenvector matrix The data point in The values ​​on the eigenvectors, Representative The weight parameter of the feature vector is used to adjust the influence of the feature vector. Represents the exponential decay adjustment coefficient, which is used to control the decay amplitude of the distance between data points. Represents the normalized adjustment coefficient, which is used to normalize the result of the eigenvector matrix. Represents an exponential decay function based on Euclidean distance, which is used to measure the distance relationship between data points in the feature space. Represents the eigenvector matrix and How related the data points are along all dimensions.

5. The intelligent measurement and calibration method according to claim 4, characterized in that: The steps for obtaining the error pattern clustering result are specifically as follows: Mapping the data points to a low-dimensional space based on the eigenvector matrix, calculating the distance between the data points in the low-dimensional space using a distance metric, and performing preliminary grouping of the data points using a clustering algorithm to generate preliminary clustering results; Analyze the preliminary clustering results, identify and mark data points with high similarity as error clusters, use statistical analysis methods to determine the data distribution characteristics of each cluster, and obtain error cluster marking results; According to the error cluster labeling results, a pattern recognition algorithm is applied to each error cluster using the formula: ； Calculate the pattern characteristics of each error cluster, comprehensively refer to the weights of the data points in the cluster, and generate the error pattern clustering results; in, Indicates The pattern characteristics of the error clusters, Represents the cluster The feature vector of the data point, For data points In cluster The weight in reflects its contribution to the cluster mode. is a smoothing parameter to avoid the denominator being zero, The error cluster The total number of data points in .

6. The intelligent measurement and calibration method according to claim 5, characterized in that: The steps of obtaining the error correction model are specifically as follows: Selecting a set of local data points of each error cluster from the error pattern clustering results, applying linear regression analysis to the data to fit a local regression function, calculating initial regression coefficients, and generating a preliminary regression model; Analyze the preliminary regression model, adjust the error value to reduce the prediction deviation according to the regression analysis results, calculate the current regression coefficient, re-evaluate the error value and optimize the adjustment coefficient, and obtain an adjusted regression model; Based on the adjusted regression model, mathematical optimization techniques are applied to iteratively optimize the error correction path using the formula: ； Generate error correction paths and build error correction models; in, represents the error correction model, is the regression coefficient for each data point, representing the weight of the point in the model, is the actual observed value, is the predicted value, Represents the model adjustment coefficient, which is used to linearly correct the deviation between the predicted value and the actual value. is the nonlinear adjustment index, represents the error correction term calculated according to the new model, Provides non-linear adjustment capability.

7. The intelligent measurement and calibration method according to claim 6, characterized in that: The steps for obtaining the error correction model update result are specifically as follows: Extracting real-time error data based on the error correction model, calculating the change range of the error point in the current state, analyzing the correlation between the data, and generating error change trend analysis results; The error change trend analysis result is called to dynamically update the parameters in the correction path, adjust the weight coefficient of the path, recalculate the correction path in combination with the current error distribution, and generate a correction path update result; Combined with the update result of the correction path, a nonlinear optimization formula is applied to the dynamically adjusted weight coefficient, using the formula: ； Calculate the dynamically adjusted weights of the correction path, and combine them with the dynamically updated weights to generate the error correction model update results; in, represents the modified path weight after dynamic adjustment, is the initial weight coefficient for each error point, is the real-time error change value, represents the smoothing parameter that limits the amplitude of nonlinear adjustments, represents the smoothing parameter that limits the range of dynamic influence, is the nonlinear gain coefficient of the error change, which is used to enhance the effect of the change amplitude on the weight adjustment.

8. An intelligent measurement and calibration system, characterized in that: According to any one of claims 1 to 7, the system comprises: The data processing module calculates the Euclidean distance or cosine similarity between each group of data points based on the instrument output data, obtains the similarity value between each pair of data points, quantifies the similarity between data points, and constructs a similarity matrix; The feature extraction module calculates the fit between each data point and other data points based on the similarity matrix, constructs a Laplace matrix, performs spectral decomposition, extracts eigenvectors, performs dimensionality reduction, obtains the distribution of data points in low-dimensional space, and generates a eigenvector matrix; The error pattern recognition module maps the data points to a low-dimensional space based on the eigenvector matrix, clusters the data points according to similarity, identifies error clusters, assigns similar data points to the same error cluster, identifies the error pattern corresponding to each cluster, and generates error pattern clustering results; The error correction module selects local data in each error cluster based on the error pattern clustering result, fits the regression function, calculates the regression coefficient and adjusts the error value, calculates the error correction path, generates the error correction model, and extracts real-time error data according to real-time feedback, calculates the error change trend, updates the correction path parameters, readjusts the weight coefficient of the correction path, dynamically adjusts the correction path, and generates the error correction model update result.

Images