Accurate error quantification and traceability analysis method and system

Abstract

The invention discloses a precise error quantification and traceability analysis method and system, and relates to the technical field of precise measurement and error analysis, and the method comprises the following steps: carrying out the layering preprocessing of original data, so as to remove abnormal values, and constructing a comprehensive multi-dimensional error quantification model; based on the multi-dimensional error quantification model, positioning a key error source with the highest contribution to a total error in a mode of combining probabilistic simulation and sensitivity analysis; and analyzing the coupling relationship between different error sources and the influence of the coupling relationship on the total error. According to the accurate error quantification and traceability analysis method provided by the invention, a layered preprocessing strategy is adopted, abnormal values in original data are effectively eliminated, a solid foundation is laid for follow-up analysis, a built multi-dimensional error quantification model comprehensively covers random errors, system errors and truncation errors, and the accuracy of the analysis is improved. The integrity and accuracy of an error quantification result are ensured, and a clear direction is provided for error control.

Description

Technical Field

[0001] This invention relates to the field of precision measurement and error analysis technology, specifically to a method and system for precise error quantification and source tracing analysis. Background Technology

[0002] In fields with stringent requirements for data accuracy, such as high-end manufacturing, aerospace measurement and control, and instrument calibration, the reliability of measurement data directly affects the stability of production processes, the accuracy of equipment operation, and the credibility of experimental results. Error quantification and source tracing, as key links in ensuring data quality, have a significant impact on the advancement of core tasks in these fields due to their technological sophistication and implementation effectiveness. As the demand for controllable data accuracy in these fields continues to increase, the industry increasingly needs analytical solutions that can integrate multiple types of errors, optimize data preprocessing, and accurately locate error sources. This is crucial for addressing challenges such as the superposition of multiple errors and complex error correlations in complex scenarios, providing scientific support for subsequent error control and system performance improvement. Building a comprehensive and efficient error analysis system has become an important direction for the development of related technologies.

[0003] Current error analysis methods in the industry still have room for improvement in practical applications: The data preprocessing stage often employs a single outlier detection strategy, which is difficult to adapt to the characteristics of data with different acquisition attributes, potentially leading to insufficient outlier removal accuracy and interference with subsequent error quantification results; some error quantification models do not comprehensively cover random errors, systematic errors, and truncation errors, making it difficult to improve the completeness of the error quantification results; simultaneously, in the error source tracing process, the analytical depth of the coupling relationships between different error sources is limited, which may affect the accuracy of locating key error sources, and some analysis procedures lack a sound closed-loop verification mechanism, making it difficult to fully guarantee the stability and reliability of the error analysis results. To address these issues, this invention proposes a precise error quantification and source tracing analysis method. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a precise error quantification and source analysis method and system. This technical solution solves the problems mentioned above, such as the common use of single outlier detection in data preprocessing, which is difficult to adapt to different collected data attributes and easily affects the accuracy of subsequent error quantification; some models cannot fully cover random, systematic, and truncation errors, resulting in insufficient quantification completeness; error source analysis of the coupling relationship of error sources is shallow, limiting the accuracy of locating key error sources; and some processes lack complete closed-loop verification, requiring improvement in the stability and reliability of the results.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A precise error quantification and source tracing analysis method includes the following steps: Hierarchical preprocessing is performed on the raw data to remove outliers, and a comprehensive multidimensional error quantification model is constructed. Hierarchical preprocessing is performed on the raw data to remove outliers, and a comprehensive multidimensional error quantification model is constructed. Based on the aforementioned multidimensional error quantification model, the key error sources that contribute the most to the total error are identified through a combination of probabilistic simulation and sensitivity analysis. Analyze the coupling relationship between different error sources and their impact on the total error; Based on the results of the above steps, the error boundary range and optimization direction are output in the closed-loop analysis process.

[0006] Preferably, the multidimensional error quantization model is a multidimensional error quantization model that includes random error, systematic error and truncation error; The key error sources that contribute the most to the total error in positioning include those that contribute the most to the positioning error using a combination of sensitivity analysis and Monte Carlo simulation. The analysis of the coupling relationship between different error sources includes introducing an error correlation matrix and resolving the coupling relationship between different error sources to complete error source tracing. The closed-loop analysis process is a quantification-tracing-verification closed-loop analysis process.

[0007] Preferably, the hierarchical preprocessing of the raw data to remove outliers includes: The stratification criteria are determined based on the collection attributes of the raw data, and the raw data is divided into multiple data layers according to the determined stratification criteria. For each data layer, an outlier detection algorithm is used to detect outliers in the data. After initially marking the detected outliers, the marked outliers are then verified a second time by combining the actual data collection scenario information corresponding to that data layer. After removing the verified outliers, the coefficient of variation of the dataset after removing outliers is calculated for each data layer. When the coefficient of variation is less than a preset threshold, the outlier removal work for that data layer is considered complete.

[0008] Preferably, the construction of a comprehensive multidimensional error quantization model includes: For random errors, statistical characteristic parameters are calculated based on the preprocessed data within the window; To address systematic errors, standard reference data corresponding to the preprocessed data is obtained, and the deviation statistics between the preprocessed data and the standard reference data are calculated. To address the truncation error, a formula for calculating the truncation error is derived theoretically. By substituting the step size and related parameters from the actual calculation process, the quantitative result of the truncation error is obtained. By integrating the quantitative indicators of random error, systematic error, and truncation error, a multidimensional error quantification model is constructed.

[0009] Preferably, the method of combining sensitivity analysis and Monte Carlo simulation to locate the key error sources with the highest contribution to error includes: Define the set of error sources to be analyzed, determine the error distribution type and distribution parameters for each error source, conduct sensitivity analysis, calculate the sensitivity coefficient of each error source, and select the error sources with higher sensitivity coefficients as the key analysis objects of Monte Carlo simulation based on the sensitivity analysis results. After all iterations are completed, the comprehensive error value obtained from the simulation is statistically analyzed, the contribution of each error source to the comprehensive error is calculated, the error sources are sorted according to the size of the contribution, and the error sources whose contribution ratio exceeds the preset ratio are identified as key error sources.

[0010] Preferably, the step of introducing an error correlation matrix to analyze the coupling relationship between different error sources includes: Determine the dimensions of the error correlation matrix, where the rows and columns of the matrix correspond to the identified error sources, and calculate the correlation coefficient between any two error sources. The calculated correlation coefficients are filled into the error correlation matrix to form a complete error correlation matrix. The error correlation matrix is ​​then analyzed to identify error source pairs with significant coupling relationships. For error source pairs with significant coupling relationships, the direction and strength of the coupling relationship are further analyzed.

[0011] Preferably, the verification step in the closed-loop analysis process of quantification-traceability-verification includes: After completing error quantification and error source tracing, obtain the validation dataset; Substitute the validation dataset into the multidimensional error quantization model to calculate the comprehensive error quantization result of the validation dataset, and at the same time identify the key error sources in the validation dataset; Compare the error quantization results of the original data with the error quantization results of the validation dataset; Compare the key error sources in the original data with the key error sources in the validation dataset; If the comparison results do not meet the preset requirements, return to the error quantization model construction step or the error tracing step, adjust the relevant parameters, and re-execute the error quantization and tracing until the verification results meet the requirements.

[0012] Preferably, the output error boundary range includes: Based on the comprehensive error value calculated by the multidimensional error quantification model, statistical analysis methods are used to determine the distribution characteristics of the error. Based on the distribution characteristics of the error and the preset confidence level, calculate the confidence interval of the error; When calculating the confidence interval, the results of the error source coupling relationship obtained from the error correlation matrix analysis are used to correct the confidence interval. The corrected confidence interval is integrated with the boundary values ​​corresponding to the error type to form a complete error boundary range output result.

[0013] Preferably, the output optimization direction includes: Based on the key error sources identified through localization, the cause of each key error source is analyzed. Based on the causes of the key error sources, formulate corresponding optimization measures; Based on the coupling relationship results obtained from the error correlation matrix analysis, the coupling relationship between error sources should be considered when formulating optimization measures.

[0014] Preferably, determining the error distribution type and distribution parameters of each error source includes: For error sources of observation equipment deviation, collect the deviation data of the equipment, use the normality test method to test the deviation data, and determine its distribution type and parameters; For error sources related to model parameter settings, their distribution type and parameters are determined based on the theoretical range of model parameters and historical adjustment records. For data transmission interference error sources, their distribution type and parameters are determined based on the interference characteristics of the transmission link; After determining the distribution parameters, the goodness-of-fit test method is used to verify the degree of fit between the error distribution type and the actual data.

[0015] This invention also provides a precise error quantification and source tracing analysis system, the system comprising a construction unit, a quantification unit, a source tracing unit, and an analysis unit. The building unit is configured to perform hierarchical preprocessing on the raw data to remove outliers and build a comprehensive multidimensional error quantification model. The quantization unit is configured to locate the key error sources that contribute the most to the total error based on the multidimensional error quantization model by combining probabilistic simulation and sensitivity analysis. The source tracing unit is configured to analyze the coupling relationships between different error sources and their impact on the total error. The analysis unit is configured to output the error boundary range and optimization direction in the closed-loop analysis process based on the results output by the above units.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: The precise error quantification and source tracing analysis method proposed in this invention employs a hierarchical preprocessing strategy to effectively eliminate outliers in the original data, laying a solid foundation for subsequent analysis. The constructed multidimensional error quantification model comprehensively covers random errors, systematic errors, and truncation errors, ensuring the integrity and accuracy of the error quantification results. By combining probabilistic simulation and sensitivity analysis, this method can accurately locate the key error sources that contribute the most to the total error, providing a clear direction for error control. The introduction of an error correlation matrix further analyzes the coupling relationship between error sources, optimizing the accuracy of error source tracing. Through a closed-loop analysis process of quantification-source tracing-verification, it not only outputs the error boundary range but also provides targeted optimization measures, offering strong technical support for improving data quality and system performance.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the precise error quantification and source tracing analysis method of the present invention; Figure 2 This is a schematic diagram of the precise error quantification and source tracing analysis system of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figure 1 As shown, Figure 1This is a schematic diagram of the precise error quantification and source tracing analysis method of the present invention. First, the raw data undergoes hierarchical preprocessing to remove outliers. Specifically, based on the data acquisition attributes, including data acquisition dimension, acquisition time period, and acquisition device type, the raw data is divided into multiple data layers. For each data layer, outlier detection algorithms are used, including the 3σ criterion based on statistical principles, box plot methods, and the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm based on cluster analysis. After initial labeling of the detected outliers, a secondary verification is performed on the labeled outliers, taking into account the actual acquisition scenario information corresponding to that data layer, such as environmental parameters and equipment operating status parameters during acquisition, to eliminate mislabeled data caused by temporary environmental interference or momentary equipment failure. After the verified outliers are removed, the coefficient of variation of the dataset after outlier removal for each data layer is calculated. When the coefficient of variation is less than the preset threshold, the outlier removal work of the data layer is considered complete; if the coefficient of variation is greater than or equal to the preset threshold, the process returns to the outlier identification step, and the parameters of the outlier identification algorithm are adjusted before outlier detection and removal are performed again.

[0022] When adjusting the parameters of an outlier detection algorithm, if the 3σ criterion is used, the original parameters are typically set at 3 times the standard deviation as the outlier threshold. When adjusting the parameters, the multiplier should be adjusted according to the dispersion of the data layer. If the dispersion of the data layer is large, the multiplier can be adjusted to 2.5 times or 2 times the standard deviation to increase the number of outliers identified. If the dispersion of the data layer is small but still exceeds the threshold, the multiplier can be adjusted to 3.5 times or 4 times the standard deviation to avoid excessively removing normal data.

[0023] If using the box plot method, the original parameters use 1.5 times the interquartile range as the outlier threshold. When adjusting the parameters, the interquartile range can be adjusted to 1 or 2 times; increase the factor when the dispersion is high and decrease the factor when the dispersion is low. If using the DBSCAN algorithm, the adjusted parameters include the neighborhood radius and the minimum number of samples. If the dispersion coefficient is too large after the initial detection, the neighborhood radius can be decreased or the minimum number of samples can be increased to reduce the number of clusters and allow more discrete data to be identified as outliers. If the large dispersion coefficient is due to normal data being misclassified as outliers, the neighborhood radius can be increased or the minimum number of samples can be decreased to include more data in the normal clusters.

[0024] After each parameter adjustment, outlier detection and removal are performed on the data layer again, and the adjusted coefficient of variation is calculated. If the preset threshold is still not met, the parameters are adjusted again until the coefficient of variation meets the requirements. At the same time, the magnitude of each parameter adjustment and the corresponding change in the coefficient of variation are recorded to form a parameter adjustment log, which provides a reference for the layered preprocessing of similar data in the future.

[0025] After completing the hierarchical preprocessing of the raw data, a comprehensive multidimensional error quantification model is constructed. This model includes the quantification of random error, systematic error, and truncation error. For random error, a time window or spatial window for data collection is determined. Based on the preprocessed data within the window, statistical characteristic parameters, such as standard deviation, variance, and root mean square error, are calculated. The calculated statistical characteristic parameters are used as quantification indicators of random error. At the same time, the size of the statistical window and the basis for its determination are recorded to reflect the statistical characteristics of random error.

[0026] In this embodiment of the invention, for systematic errors, standard reference data corresponding to the preprocessed data is obtained, and the mean deviation, maximum deviation, and deviation trend between the preprocessed data and the standard reference data are calculated. A functional relationship between systematic errors and data acquisition variables is established through linear regression or nonlinear fitting methods. This functional relationship and the corresponding deviation statistics are used as the quantification content of systematic errors. Data acquisition variables include acquisition time, acquisition location, and operating parameters of the acquisition device.

[0027] Specifically, regarding truncation error, the numerical calculation method used in data processing or model calculation is identified, the truncation order of this numerical calculation method is determined, and the calculation formula for truncation error is derived theoretically by combining the step size parameter in the calculation process. Substituting the step size value and related parameters from the actual calculation process, the quantitative result of truncation error is obtained. The quantitative indicators and related calculation basis of random error, systematic error, and truncation error are integrated to construct a multi-dimensional error quantification model. The model includes an error weight allocation module. Based on the degree of influence of different error types on the final data or model results, the weight of each error type is determined using the analytic hierarchy process (AHP) to achieve a quantitative assessment of the overall error.

[0028] Based on the constructed multidimensional error quantification model, the key error sources that contribute the most to the total error are located by combining probabilistic simulation and sensitivity analysis.

[0029] The set of error sources to be analyzed is clearly defined. This set is determined based on the data acquisition and model building stages. In addition to observation equipment bias and model parameter settings, it also includes interference error sources during data transmission and truncation error sources during data preprocessing. For each error source, its error distribution type is determined. Error distribution types include normal distribution, uniform distribution, and Poisson distribution. Based on historical data or theoretical analysis, the distribution parameters of each error source are determined, such as mean, standard deviation, and variance.

[0030] For error sources related to observation equipment deviation, deviation data of the equipment under different time periods and operating conditions are collected. Normality tests are then performed on the deviation data, including the Shapiro-Wilk test and the Kolmogorov-Smirnov test. If the test results indicate that the deviation data conforms to a normal distribution, the mean and standard deviation of the data are calculated as parameters of the normal distribution. If it does not conform to a normal distribution, the distribution pattern of the data is further analyzed to determine whether it conforms to a uniform distribution or other distribution types. For deviation data conforming to a uniform distribution, the range of values ​​is calculated, and the lower and upper limits of the range are used as parameters of the uniform distribution. For error sources related to model parameter setting, based on the theoretical range of model parameters and historical adjustment records, the possible value intervals of the parameters are determined. If the parameter values ​​are uniformly distributed within the interval, a uniform distribution is used to describe its error distribution, with the upper and lower limits of the value interval used as distribution parameters. If the parameter values ​​fluctuate around an optimal value, and the fluctuation amplitude conforms to the characteristics of a normal distribution, a normal distribution is used, with the optimal value as the mean and the statistical value of the fluctuation amplitude as the standard deviation.

[0031] For data transmission interference-related error sources, based on the interference characteristics of the transmission link, if the occurrence of interference events conforms to a Poisson distribution, then a Poisson distribution is used to describe its error distribution, with the average occurrence rate as the distribution parameter. After determining the distribution parameters, a goodness-of-fit test is used to verify the degree of fit between the error distribution type and the actual data. Goodness-of-fit tests include the chi-square test and the Anderson-Darling test. When the goodness of fit is greater than a preset fitting threshold, the determined error distribution type and distribution parameters are deemed reasonable; otherwise, the data is reanalyzed and the distribution type and parameters are adjusted.

[0032] Sensitivity analysis is conducted using the local sensitivity analysis method. The values ​​of other error sources are fixed, while the value of the target error source is changed. The change in the final data or model results caused by the change in the value of the target error source is calculated, and the sensitivity coefficient of the target error source is obtained. The sensitivity coefficient is used to preliminarily determine the degree of influence of each error source on the results.

[0033] Based on the sensitivity analysis results, error sources with high sensitivity coefficients are selected as the key analysis objects of Monte Carlo simulation. The number of iterations of Monte Carlo simulation is set according to the calculation accuracy requirements. In each iteration, the values ​​of error sources are randomly generated according to the error distribution type of each error source, and substituted into the multidimensional error quantification model to calculate the corresponding comprehensive error value.

[0034] After all iterations are completed, the comprehensive error value obtained from the simulation is statistically analyzed. The contribution of each error source to the comprehensive error is calculated in all iterations. The contribution is calculated by the covariance between the change in the value of the error source and the change in the comprehensive error. The error sources are sorted according to the size of the contribution. The error sources with a contribution ratio exceeding a preset ratio are identified as key error sources. The preset ratio is set according to the accuracy requirements of error control in the actual application scenario.

[0035] To analyze the coupling relationships between different error sources and their impact on the total error, an error correlation matrix is ​​introduced to analyze these relationships. First, the dimensions of the error correlation matrix are determined, with rows and columns corresponding to identified error sources, and matrix elements representing the correlation coefficient between two corresponding error sources. Based on historically collected error source data and corresponding comprehensive error data, the correlation coefficient between any two error sources is calculated using methods including Pearson correlation coefficient and Spearman rank correlation coefficient. The Pearson correlation coefficient method is used when the error source data follows a normal distribution, while the Spearman rank correlation coefficient method is used when the data does not follow a normal distribution.

[0036] The calculated correlation coefficients are filled into the error correlation matrix to form a complete error correlation matrix. The error correlation matrix is ​​analyzed, and when the absolute value of the correlation coefficients corresponding to two error sources in the matrix is ​​greater than a preset correlation threshold, it is determined that the two error sources have a significant coupling relationship. The preset correlation threshold is determined based on the confidence level of the error analysis.

[0037] For error source pairs exhibiting significant coupling, the direction and strength of this coupling are further analyzed. A coupling function is constructed, using the error values ​​of the two error sources as input variables and the combined error value as the output variable. Multiple regression analysis is employed to determine the coefficients of the coupling function, quantifying the degree of coupling. Based on the coupling analysis results, the identification of key error sources is adjusted. If one of two significantly coupled error sources is initially identified as a key error source, the other error source must also be included in the category of key error sources to ensure the comprehensiveness of subsequent error control and optimization.

[0038] Based on the results of the above steps, an error boundary range and optimization direction are output in a closed-loop analysis process of "quantification-source tracing-verification" to improve the reliability of data or model results. After completing error quantification and source tracing, a verification dataset is obtained. The verification dataset and the original data belong to the same data type, but the collection time or collection conditions differ from the original data to ensure the objectivity of the verification results. The verification dataset is substituted into the multidimensional error quantification model to calculate the comprehensive error quantification result of the verification dataset. At the same time, sensitivity analysis and Monte Carlo simulation are used to identify the key error sources in the verification dataset. The error quantification results of the original data and the verification dataset are compared to calculate the error deviation rate between the two. When the error deviation rate is less than the preset deviation threshold, the stability of the error quantification model is deemed to meet the requirements.

[0039] Compare the key error sources in the original data with those in the validation dataset, and calculate their overlap. If the overlap is greater than a preset overlap threshold, the consistency of the error tracing results is considered to meet the requirements. If the error deviation rate is greater than or equal to a preset deviation threshold, or the overlap of key error sources is less than or equal to a preset overlap threshold, return to the error quantization model construction step or the error tracing step to check for problems in the model parameter settings, error source distribution parameter determination, etc. Adjust the relevant parameters and re-execute error quantization and tracing until the validation results meet the deviation threshold and overlap threshold requirements, thus completing the closed-loop validation.

[0040] When outputting the error boundary range, the comprehensive error value calculated based on the multidimensional error quantification model is used to determine the error distribution characteristics using statistical analysis methods. These characteristics include distribution type, mean, and standard deviation. Based on the error distribution characteristics and a preset confidence level, the error confidence interval is calculated. This confidence interval is the core content of the error boundary range, and the confidence level is determined according to the reliability requirements of the results in the actual application scenario. When calculating the confidence interval, the error source coupling relationship results obtained from the error correlation matrix analysis are used to correct for error fluctuations caused by error source coupling. The correction method involves adjusting the width of the confidence interval according to the strength of the coupling relationship. When a significant coupling relationship exists, the width of the confidence interval is appropriately increased to ensure that the error boundary range can cover the actual possible error values. The corrected confidence interval is integrated with the boundary values ​​corresponding to the error type to form a complete error boundary range output result. The boundary value for each error type is determined by its respective quantification index: the boundary value for random error is twice the standard deviation of its statistical characteristic parameter; the boundary value for systematic error is the maximum deviation; and the boundary value for truncated error is the theoretically derived upper limit of the error.

[0041] When outputting optimization directions, the causes of each key error source are analyzed based on the identified location. For key error sources related to observation equipment deviation, the causes include excessively long equipment calibration cycles, aging equipment hardware, and deviations in equipment installation location. For key error sources related to model parameter settings, the causes include parameter values ​​being set based on experience without data support, and parameters not being dynamically adjusted according to changes in data acquisition conditions. Based on the causes of the key error sources, corresponding optimization measures are formulated. For deviations caused by excessively long equipment calibration cycles, the optimization measure is to shorten the calibration cycle, specifying that the new calibration cycle length should be determined based on the changing trends of historical deviation data. For deviations caused by aging equipment hardware, the optimization measure is to replace aging hardware components, specifying that the criterion for component replacement is the percentage by which hardware performance parameters fall below the preset factory standard. For errors caused by unreasonable model parameter settings, the optimization measure is to re-determine parameter values ​​based on preprocessed data using parameter optimization algorithms, including gradient descent and genetic algorithms.

[0042] In this embodiment of the invention, the coupling relationship results obtained by combining the error correlation matrix analysis are considered when formulating optimization measures. If there is a significant coupling relationship between two key error sources, optimization measures for the two error sources need to be formulated simultaneously to avoid the error of the other error source being increased due to the optimization of one error source alone. This ensures that the optimization measures can comprehensively reduce the overall error and improve the reliability of the data or model results.

[0043] This invention also provides a precise error quantification and source tracing analysis system. Figure 2 This is a schematic diagram of the precise error quantification and source tracing analysis system of the present invention. Figure 2 The system comprises a construction unit, a quantization unit, a source tracing unit, and an analysis unit. The construction unit is configured to perform hierarchical preprocessing on the raw data to remove outliers and construct a comprehensive multidimensional error quantization model. The quantization unit is configured to locate the key error sources that contribute the most to the total error based on the multidimensional error quantization model through a combination of probabilistic simulation and sensitivity analysis. The source tracing unit is configured to analyze the coupling relationship between different error sources and their impact on the total error. The analysis unit is configured to output the error boundary range and optimization direction in a closed-loop analysis process based on the results output by the above units.

[0044] The precise error quantification and source tracing analysis method proposed in this invention employs a hierarchical preprocessing strategy to effectively eliminate outliers in the original data, laying a solid foundation for subsequent analysis. The constructed multidimensional error quantification model comprehensively covers random errors, systematic errors, and truncation errors, ensuring the integrity and accuracy of the error quantification results. By combining probabilistic simulation and sensitivity analysis, this method can accurately locate the key error sources that contribute the most to the total error, providing a clear direction for error control. The introduction of an error correlation matrix further analyzes the coupling relationship between error sources, optimizing the accuracy of error source tracing. Through a closed-loop analysis process of quantification-source tracing-verification, it not only outputs the error boundary range but also provides targeted optimization measures, offering strong technical support for improving data quality and system performance.

[0045] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of precision error quantification and traceability analysis, characterized in that, The method comprises the following steps: performing hierarchical preprocessing on the original data to eliminate outliers therein, and constructing a comprehensive multi-dimensional error quantification model; based on the multi-dimensional error quantification model, locating the key error source with the highest contribution to the total error by combining probabilistic simulation and sensitivity analysis; analyzing the coupling relationship between different error sources and their influence on the total error; based on the results of the above steps, outputting the error boundary range and optimization direction in a closed-loop analysis process.

2. The method according to claim 1, characterized in that: the multi-dimensional error quantification model is a multi-dimensional error quantification model comprising random error, systematic error and truncation error; the locating of the key error source with the highest contribution to the total error comprises locating the key error source with the highest contribution ratio by combining sensitivity analysis and Monte Carlo simulation; the analysis of the coupling relationship between different error sources comprises introducing an error correlation matrix to analyze the coupling relationship between different error sources and complete error tracing; the closed-loop analysis process is a closed-loop analysis process of quantification-tracing-verification.

3. The method of claim 1, wherein, The hierarchical preprocessing on the original data to eliminate outliers therein comprises: determining hierarchical basis according to the collection attributes of the original data, and dividing the original data into multiple data layers according to the determined hierarchical basis; for each data layer, respectively detecting outliers in the data by using an outlier identification algorithm; after preliminarily marking the detected outliers, re-verifying the marked outliers in combination with the actual collection scene information corresponding to the data layer; after eliminating the verified outliers, calculating the dispersion coefficient of the data set after eliminating outliers for each data layer, and determining that the outlier elimination work for the data layer is completed when the dispersion coefficient is less than a preset threshold.

4. The method of claim 1, wherein, The construction of the comprehensive multi-dimensional error quantification model comprises: for random error, calculating statistical characteristic parameters based on the preprocessed data within a window; for systematic error, obtaining standard reference data corresponding to the preprocessed data, and calculating the deviation statistics between the preprocessed data and the standard reference data; for truncation error, deriving a calculation formula of the truncation error through theoretical derivation, substituting the step value and related parameters in the actual calculation process into the formula, and obtaining the quantification result of the truncation error; integrating the quantification indexes of random error, systematic error and truncation error to construct a multi-dimensional error quantification model.

5. The method of precise error quantification and traceability analysis of claim 2, wherein, The locating of the key error source with the highest contribution ratio by combining sensitivity analysis and Monte Carlo simulation comprises: clearly defining the error source set to be analyzed, determining the error distribution type and distribution parameters of each error source, carrying out sensitivity analysis, calculating the sensitivity coefficients of the error sources, and selecting the error sources with higher sensitivity coefficients as the key analysis objects of Monte Carlo simulation based on the sensitivity analysis results; after completing all iterations, statistically analyzing the comprehensive error values obtained by simulation, calculating the contribution of each error source to the comprehensive error, sorting the error sources according to the contribution, and determining the error sources with a contribution ratio exceeding a preset proportion as the key error sources.

6. The method of precise error quantification and traceability analysis of claim 2, wherein, The introduction error correlation matrix analyzes the coupling relationship between different error sources, including: Determine the dimension of the error correlation matrix, the rows and columns of the matrix correspond to the identified error sources, and calculate the correlation coefficient between any two error sources; Fill the calculated correlation coefficient into the error correlation matrix to form a complete error correlation matrix, analyze the error correlation matrix, and determine the error source pair with significant coupling relationship; Further analyze the direction and strength of the coupling relationship for the error source pair with significant coupling relationship.

7. The method of precise error quantification and traceability analysis of claim 2, wherein, The verification step in the closed-loop analysis process of the quantification-tracing-verification includes: After completing error quantification and error tracing, obtain the verification data set; Substitute the verification data set into the multi-dimensional error quantification model to calculate the comprehensive error quantification result of the verification data set and determine the key error sources in the verification data set; Compare the error quantification results of the original data and the verification data set; Compare the key error sources of the original data and the verification data set; If the comparison result does not meet the preset requirement, return to the error quantification model construction step or the error tracing step, adjust the relevant parameters and re-execute the error quantification and tracing until the verification result meets the requirement.

8. The method of precise error quantification and traceability analysis of claim 1, wherein, The output error boundary range includes: Based on the comprehensive error value calculated by the multi-dimensional error quantification model, determine the distribution characteristics of the error using statistical analysis method; According to the distribution characteristics of the error and the preset confidence level, calculate the confidence interval of the error; When calculating the confidence interval, combine the error source coupling relationship result obtained by analyzing the error correlation matrix to correct the confidence interval; Integrate the corrected confidence interval and the boundary value corresponding to the error type to form a complete error boundary range output result.

9. The method of precise error quantification and traceability analysis of claim 1, wherein, The output optimization direction includes: Based on the key error sources obtained by positioning, analyze the causes of each key error source; According to the causes of the key error sources, formulate corresponding optimization measures; Combine the coupling relationship result obtained by analyzing the error correlation matrix to consider the coupling relationship between error sources when formulating optimization measures.

10. The precision error quantification and provenance analysis method of claim 5, wherein, The determination of the error distribution type and distribution parameter of each error source includes: For observation equipment bias error sources, collect the bias data of the equipment, use normality test method to test the bias data, and determine the distribution type and parameter; For model parameter setting error sources, determine the distribution type and parameter based on the theoretical value range of the model parameters and the historical adjustment records; For data transmission interference error sources, determine the distribution type and parameter according to the interference characteristics of the transmission link; After determining the distribution parameters, use goodness-of-fit test method to verify the fitting degree of the error distribution type and the actual data.

11. A precision error quantification and traceability analysis system, characterized in that, The system includes a construction unit, a quantification unit, a tracing unit and an analysis unit, The construction unit is configured to perform hierarchical preprocessing on the original data to eliminate outliers therein and construct a comprehensive multi-dimensional error quantification model; The quantification unit is configured to locate the key error sources with the highest contribution to the total error based on the multi-dimensional error quantification model by combining probability simulation and sensitivity analysis. a traceability unit configured to analyze the coupling relationship between different error sources and their influence on the total error; an analysis unit configured to output the error boundary range and the optimization direction in a closed-loop analysis process based on the results output by the above units.

Images