A deep learning-based metrological detection data automatic identification method

Abstract

The present application relates to the technical field of metering data, in particular to a kind of metering detection data automatic identification method based on deep learning, in the present application, signal strength score mark distortion section, signal distortion is found in time in data acquisition stage, and distortion signal is corrected and reconstructed by wavelet transform, so that the accuracy of original data is improved, the precision of abnormal mode in data is realized by generating adversarial network Precision mining, simultaneously associated abnormal reason and form causal report, multiple regression analysis starts from error source, quantifies the intensity of each influencing factor, provides clear direction and basis for subsequent data optimization, through the classification analysis capability of support vector machine, the characteristics of the corrected data are compared and accurately classified step by step, and the classification mark result is further layered and arranged by deep learning model, improve the reliability, accuracy and efficiency of metering data in transmission, storage and analysis process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metrology data, and particularly relates to a metrology detection data automatic identification method based on deep learning. BACKGROUND

[0002] The technical field of metrology data involves accurate measurement of various physical quantities, and processing, analysis and management of the measured data, with the goal of obtaining accurate data through measurement equipment or systems, ensuring consistency and reliability in transmission, storage and processing.

[0003] The metrology detection data automatic identification method based on deep learning aims to realize intelligent identification, classification and processing of complex data in the metrology detection process, realize automatic analysis of large-scale metrology data through the introduction of deep learning technology, improve the accuracy and reliability of detection, and reduce the need for manual intervention, so as to improve the detection efficiency and optimize the quality control process. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and propose a metrology detection data automatic identification method based on deep learning.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: a metrology detection data automatic identification method based on deep learning, comprising the following steps:

[0006] Step one: receiving raw data through a sensor, and scoring signal strength, marking and recording distortion segments to generate a signal quality evaluation report;

[0007] Step two: according to the signal quality evaluation report, using wavelet transform to correct the marked segments and correct the distorted signals to obtain reconstructed signal data;

[0008] Step three: structure decomposition is performed on the reconstructed signal data, abnormal patterns and abnormal reasons are identified through a generative adversarial network, the generative adversarial network is continuously trained to obtain a deep learning model, and the relationship between abnormal patterns and abnormal reasons is recorded to generate an abnormal pattern and causal association report;

[0009] Step four: using the abnormal pattern and causal association report, the influence intensity of each factor is calculated through multivariate regression analysis, and the error source is evaluated to obtain error source quantitative analysis results;

[0010] Step five: according to the error source quantitative analysis results, the original data is corrected by accurately adjusting parameters to optimize the accuracy and reliability of the original data, and inverse corrected data is generated;

[0011] Step six: based on the reverse corrected data, the feature range of each data is extracted, the data is added with a label value according to a classification standard, and a support vector machine is used to compare and analyze the reverse corrected data and the classification feature value, and output the classification result, and the classification label of each item is input into a deep learning model to obtain the measurement data category recognition result.

[0012] As a further scheme of the present application, the specific steps for generating the signal quality evaluation report are:

[0013] Based on the original data received by the sensor, the intensity value of the signal data is extracted one by one, the intensity value of each signal segment is recorded by the segmented scanning detection method, the signal intensity is quantitatively scored and archived according to the preset scoring rule, the starting and ending positions of all intensity abnormal segments are marked, and signal scoring and distortion segment marking data are generated;

[0014] Based on the signal scoring and distortion segment marking data, by traversing all scoring and marking records, each signal segment and its corresponding scoring and marking information are integrated, the number and length of the distortion segments are statistically classified, the overall coverage ratio of the distortion signal is calculated by aggregation, and scoring statistical data and segment marking information are generated;

[0015] Based on the scoring statistical data and segment marking information, the scoring information of each signal segment is filtered to comprehensively analyze the coverage range and quality distribution of the signal, the distribution ratio and quality score of the distortion signal are normalized calculated, and a signal quality evaluation report is generated;

[0016] The signal quality evaluation report includes the specific position of the distortion segment, the signal intensity level and the noise level.

[0017] As a further scheme of the present application, the specific steps for generating the reconstructed signal data are:

[0018] Based on the signal quality evaluation report, the distortion segment signal data is filtered, the distortion signal is cut into fixed length by segment division operation, the numerical change range of each signal segment is analyzed and calculated, the characteristic parameter table of each segment is generated by calculating the fluctuation rate and abnormal amplitude, and distortion signal segment characteristic data is generated;

[0019] Based on the distortion signal segment characteristic data, the segment signal is processed by segment traversal, the abnormal jump value appearing in the signal is interpolated, the signal change characteristic is corrected by adjusting the continuity of the signal boundary value, and the signal distribution is numerically equalized, and segment signal correction data is generated.

[0020] Based on the segment signal correction data, the corrected signal of each segment is spliced with normal signal data in original order, the data characteristics of the splicing boundary are analyzed point by point, the connection value and fluctuation rate between all segments are processed, the overall consistency between signal segments is adjusted, and the reconstructed signal data is generated.

[0021] The reconstructed signal data includes reconstructed waveform, adjusted frequency response and signal stability score.

[0022] As a further scheme of the application, the specific steps for generating the abnormal pattern and causal correlation report are:

[0023] Based on the reconstructed signal data, the signal data is divided into multiple independent hierarchical data regions according to time range and signal amplitude change trend, the change rate and structural characteristics of the signal are extracted by layer-by-layer scanning, the key parameters of the signal are extracted, and the abnormal data part existing in the signal is separated, the dynamic characteristics separation of abnormal data is completed, and the signal decomposition feature data is generated.

[0024] Based on the signal decomposition feature data, the latent distribution characteristics of the signal decomposition feature data are learned by using the generator of the generative adversarial network, the generated data and the actual data are compared, the abnormal pattern characteristics in the signal distribution are captured, the abnormal pattern characteristics are extracted as key dynamic parameters, the abnormal pattern characteristics are matched with known characteristics of potential abnormal causes, the correlation mapping of the pattern and the cause is completed, the analysis results are arranged as a multi-dimensional parameter set, and the abnormal pattern and cause relationship data are generated.

[0025] Based on the abnormal pattern and cause relationship data, the causal characteristics of the abnormal pattern are modeled by using the generative adversarial network, the causal characteristic distribution is reconstructed, the consistency between the generated distribution and the actual distribution is evaluated, the dynamic correlation path of the causal variable is determined, the variables in the causal path are analyzed item by item by using the feature relationship mapping method, the causal strength between the pattern characteristics and the abnormal cause is quantified, the modeling results are recorded as the causal correlation data set, the causal relationship model is obtained, and the abnormal pattern and causal correlation report is generated.

[0026] The abnormal pattern and causal correlation report includes the list of identified abnormal patterns, related cause analysis and impact degree evaluation.

[0027] As a further scheme of the present application, the data structure is processed by using a multi-level feature decomposition method to extract feature parameters in each layer of data, the signal data is preliminarily layered according to time periods and frequency ranges, the data is divided into multiple different characteristic intervals, the amplitude, change rate and structure parameters of the layered data are extracted by layer-by-layer scanning, the dynamic characteristics of each layer of data are grouped and aggregated, the change range and change trend of the fluctuation amplitude are extracted, the extracted dynamic characteristics are compared with abnormal parameters layer by layer, the change range of abnormal characteristics is recorded, and the abnormal signal content with dynamic characteristics is separated out.

[0028] The mode characteristics in the abnormal signal are identified, the mode characteristics are matched with potential abnormal causes, the dynamic characteristics related to time series changes in the data are extracted, the amplitude difference, periodic change and signal mutation point in different time points and frequency ranges are compared, the characteristic distribution of the data is itemized filtered, the abnormal signal characteristic range is found, the fluctuation mode is further analyzed, the repeated change period, peak interval and change law are extracted, the existing characteristic parameter set is matched, the change mode of the potential abnormal cause is compared, and the specific association information between the mode and the cause is recorded.

[0029] As a further scheme of the present application, the specific steps for generating the error source quantitative analysis result are:

[0030] Based on the abnormal mode and causal association report, the causal factor information in the report is extracted, the weight analysis of the multivariate relationship is performed, the association strength and weight parameter between the multivariables are decomposed, and a causal influence weight distribution result is generated.

[0031] Based on the causal influence weight distribution result, the error data in the variable is extracted, the statistical analysis of the error bias is performed, the error source is evaluated by parameter difference decomposition, and an error source statistical result is generated.

[0032] Based on the error source statistical result, the error source is quantitatively decomposed, the error distribution characteristics of the causal variables are extracted, the quantitative evaluation is completed by combining the error weight calculation, and an error source quantitative analysis result is generated.

[0033] The error source quantitative analysis result includes a main error source list, a contribution rate of each source and an environmental factor affecting the error.

[0034] As a further scheme of the present application, the specific steps for generating the data after the reverse correction are:

[0035] Based on the error source quantitative analysis result, the specific parameters in the error distribution are extracted, each parameter is grouped and analyzed according to the bias range, the bias value of each group of parameters is calculated, the data range in the group is adjusted, a correction bias parameter set is generated.

[0036] Based on the correction deviation parameter set, each parameter in the original data is corrected one by one, the correction value in the grouping range is used to replace the data, the deviation value is corrected by the cumulative way, and the corrected original data is generated;

[0037] Based on the corrected original data, residual abnormal values in the data that are not corrected are extracted, out-of-range deviation evaluation is performed on the abnormal values, the deviation parameters of the abnormal values are gradually adjusted, and the inversely corrected data is generated;

[0038] The inversely corrected data includes adjusted data parameters, data comparison before and after optimization, and correction effect evaluation.

[0039] As a further scheme of the present application, the specific steps for generating the measurement data category recognition result are:

[0040] Based on the inversely corrected data, the classification feature values in the data are scanned item by item, the feature range of each item of data is extracted, a label value is added to the data according to the classification standard, and a data classification feature set is generated;

[0041] Based on the data classification feature set, the inversely corrected data is compared and analyzed with the classification feature values by using a support vector machine, the classification label of each item of data is gradually confirmed, the data is classified and arranged in combination with the label result, and a data classification label result is generated;

[0042] Based on the data classification label result, each classification label is input into a deep learning model, and hierarchical arrangement of classification output is performed in combination with the input data, the classified measurement data is extracted, and a measurement data category recognition result is generated;

[0043] The measurement data category recognition result includes the classified data type, the classification accuracy, and the decision basis of the model.

[0044] As a further scheme of the present application, the support vector machine is according to the formula:

[0045]

[0046] Wherein: f'(x) is the classification label result of the measurement detection data, x is the feature vector of the measurement detection data to be identified, x i is the feature vector of the support vector that has been trained in the support vector machine, alpha i is the Lagrange multiplier weight of the support vector, y i is the category label corresponding to the support vector, exp(-gamma||x i -x|| 2 ) is a Gaussian kernel function, gamma is a scale parameter of the Gaussian kernel, theta is a dynamic adjustment parameter, and w is a feature weighting vector. is the characteristic weighted result of the linear part of the kernel function, b is the bias term of the model, and lambda||x|| is the regularization term. 2 is the regularization term, and lambda is the regularization coefficient.

[0047] As a further scheme of the present application, the classification feature values in the data are scanned item by item, and the feature range of each item of data is extracted, the classification feature values being attributes and indexes used for classifying categories in the data, and the extraction step specifically comprises: from the data after reverse correction, numerical type and category type attributes are extracted according to a predefined classification rule, for the numerical type attribute, a value is directly extracted, and for the category type attribute, a numerical code is converted by using a dictionary mapping, for missing values, interpolation and default value filling are adopted, and finally the feature values are sorted and classified.

[0048] Compared with the prior art, the present application has the advantages and positive effects that:

[0049] 1. In the present application, the distorted section is marked by signal intensity score, the signal distortion is found in time in the data acquisition stage, and the distorted signal is corrected and reconstructed by wavelet transform, so that the accuracy of the original data is improved.

[0050] 2. In the present application, the generation of the adversarial network realizes the accurate mining of the abnormal mode in the data, simultaneously associates the abnormal causes and forms a causal report, and the multivariate regression analysis starts from the error source and quantifies the strength of each influencing factor, so as to provide a clear direction and basis for subsequent data optimization.

[0051] 3. In the present application, the classification analysis ability of the support vector machine is used to gradually compare and accurately classify the features of the corrected data, so that the classification result is more in line with the law characteristics of the data itself, the reliability and accuracy of the classification are improved, and the classification marked result is further layered and sorted by the deep learning model, so that the data output by the classification is more hierarchical and orderly, and the reliability, accuracy and efficiency of the measurement data in the transmission, storage and analysis process are improved. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 is a main step schematic diagram of the present application;

[0053] Figure 2 is a S1 refinement schematic diagram of the present application;

[0054] Figure 3 is a S2 refinement schematic diagram of the present application;

[0055] Figure 4 is a S3 refinement schematic diagram of the present application;

[0056] Figure 5 is a S4 refinement schematic diagram of the present application;

[0057] Figure 6 S5 refinement schematic diagram of the present application;

[0058] Figure 7 S6 refinement schematic diagram of the present application. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0060] Please refer to Figure 1 The present application provides a technical scheme: a metrological detection data automatic identification method based on deep learning, comprising the following steps:

[0061] S1: receiving original data through a sensor, scoring through signal strength, marking and recording distortion sections, and generating a signal quality evaluation report;

[0062] S2: according to the signal quality evaluation report, using wavelet transform to correct the marked sections, and correcting the distorted signals to obtain reconstructed signal data;

[0063] S3: structure decomposition is performed on the reconstructed signal data, abnormal patterns and abnormal reasons are identified through a generative adversarial network, the generative adversarial network is continuously trained to obtain a deep learning model, and the relationship between the abnormal patterns and the abnormal reasons is recorded to generate an abnormal pattern and causal association report;

[0064] S4: using the abnormal pattern and causal association report, through multivariate regression analysis, the influence intensity of each factor is calculated, and the error source is evaluated to obtain error source quantitative analysis results;

[0065] S5: according to the error source quantitative analysis results, the original data is corrected by accurately adjusting parameters, the accuracy and reliability of the original data are optimized, and inverse corrected data is generated;

[0066] S6: based on the inverse corrected data, the feature range of each data is extracted, a label value is added to the data according to a classification standard, and a support vector machine is used to compare and analyze the inverse corrected data and the classification feature value, output the classification result, input each classification label into the deep learning model, and obtain the metrological data category identification result.

[0067] Please refer to Figure 2 The specific steps for generating the signal quality evaluation report are:

[0068] S101: Based on the raw data received by the sensor, the intensity value of the signal data is extracted one by one, the intensity value of each signal segment is recorded by the segmented scanning detection method, the signal intensity is quantitatively scored according to the preset scoring rule and is archived, the starting and ending positions of all intensity abnormal segments are marked, and signal scoring and distortion segment marking data are generated;

[0069] S102: Based on the signal scoring and distortion segment marking data, by traversing all scoring and marking records, each signal segment and its corresponding scoring and marking information are integrated, the number and length of distortion segments are statistically classified, the overall coverage ratio of distortion signals is calculated by aggregation, and scoring statistical data and segment marking information are generated;

[0070] S103: Based on the scoring statistical data and segment marking information, the scoring information of each signal segment is filtered to comprehensively analyze the coverage range and quality distribution of the signal, and the distribution ratio and quality score of the distortion signal are normalized to generate a signal quality evaluation report;

[0071] Based on the raw data received by the sensor, a segmented scanning detection method is used to divide the signal data into fixed time windows, the window size is 5 milliseconds, the signal intensity value is calculated in each signal segment, the sliding window technology is used, the window size is also 5 milliseconds, and the step length is 1 millisecond. The continuity of the signal is detected, and the detection value is saved in array form. By setting a threshold value of 0.2, the intensity value of each signal segment is judged and quantitatively scored. According to the scoring rule, the score value is archived as an integer type value, ranging from 0 to 100. The signal segment with an intensity value less than the threshold value is marked and its starting and ending positions are recorded, and signal scoring and distortion segment marking data are generated.

[0072] Based on the signal scoring and distortion segment marking data, by traversing all scoring records and segment markings, a statistical classification algorithm is used to classify and process the signal scoring, which is divided into five levels: 0-20, 21-40, 41-60, 61-80, and 81-100. The corresponding relationship between signal scoring and distortion segments is mapped, a segment merging algorithm is used, and merging is performed according to the starting and ending positions of the segments. The merging condition is that the interval between two segments is less than 2 milliseconds. The length and number of merged segments are counted, and the distortion coverage rate is defined as the ratio of the total length of distortion segments to the total length of signals using aggregation calculation. All statistical results are stored in the form of a dictionary, and scoring statistical data and segment marking information are generated.

[0073] Based on the score statistical data and the section mark information, the numerical distribution range of the signal score is screened, the interval statistics of the score value is performed by using a histogram calculation algorithm, the score value is divided into 10 equal interval ranges, the statistical quantity of each interval range is stored in a list, the normalized score is defined as the original score value divided by 100 in a normalized manner, the coverage is defined as the ratio of the signal length in a specific score range to the total signal length by using a distribution proportion calculation method, the signal quality comprehensive score is defined as the weighted sum of the coverage of the score interval by using a comprehensive analysis algorithm, the weight is set as the upper limit value of the score interval divided by 100, the distribution histogram of the signal score and the signal coverage range diagram are generated by using a statistical function, and a signal quality evaluation report is generated.

[0074] The signal quality evaluation report includes the specific position of the distortion section, the signal intensity level and the noise level.

[0075] Please refer to Figure 3 The specific steps of generating the reconstructed signal data are as follows:

[0076] S201: Based on the signal quality evaluation report, the distortion section signal data is screened, the fixed length cutting algorithm is used to perform the segment division operation on the distortion signal, the cutting length is set as 500 sampling points, the signal is divided and stored in an array in a fixed length by using a cutting function, the numerical value range of each signal is calculated by using an analysis algorithm, the numerical value range is defined as the difference between the maximum value and the minimum value in the segment by using a maximum difference calculation formula, the fluctuation rate of each signal is calculated according to the formula, the fluctuation rate is defined as the ratio of the standard deviation to the average value in the segment, the deviation value range in the segment is recorded by using an abnormal amplitude statistical method, and the statistical value is stored in a two-dimensional characteristic table to generate the distortion signal section characteristic data.

[0077] S202: Based on the distortion signal section characteristic data, the section signal is processed in a segment-by-segment traversal manner, the abnormal jump value appearing in the signal is supplemented by interpolation, the signal change characteristic is modified by the continuity adjustment of the signal boundary value, the signal distribution is balanced in a numerical value manner, and section signal correction data is generated.

[0078] S203: Based on the section signal correction data, each section of the corrected signal and the normal signal data are spliced in the original order, the data characteristics of the splicing boundary are analyzed point by point, the connection value and the fluctuation rate between all sections are processed, the overall consistency between the signal sections is adjusted, and the reconstructed signal data is generated.

[0079] Based on the signal quality evaluation report, the distortion section signal data is screened, the fixed length cutting algorithm is used to perform the segment division operation on the distortion signal, the cutting length is set as 500 sampling points, the signal is divided and stored in an array in a fixed length by using a cutting function, the numerical value range of each signal is calculated by using an analysis algorithm, the numerical value range is defined as the difference between the maximum value and the minimum value in the segment by using a maximum difference calculation formula, the fluctuation rate of each signal is calculated according to the formula, the fluctuation rate is defined as the ratio of the standard deviation to the average value in the segment, the deviation value range in the segment is recorded by using an abnormal amplitude statistical method, and the statistical value is stored in a two-dimensional characteristic table to generate the distortion signal section characteristic data.

[0080] Based on the distortion signal segment characteristic data, the segment signal is traversed segment by segment, the abnormal jump value in the signal is processed by using an interpolation algorithm, the identification of the abnormal jump value is completed by setting a threshold, the threshold is defined as the average value plus or minus twice the standard deviation, the values outside the range are supplemented by using a linear interpolation method, the interpolation calculation formula is defined as the linear weighted sum of adjacent valid points, the signal starting and ending positions are smoothed by using a signal boundary value continuity adjustment algorithm, the numerical distribution of the boundary points is adjusted by a convolution algorithm, the convolution window size is 3 sampling points, the distribution in the signal is processed by an equalization algorithm, the equalization processing is completed by a segmented normalization formula, the in-segment mean value is set to 0 and the standard deviation is set to 1, and segment signal correction data is generated;

[0081] Based on the segment signal correction data, each segment of the corrected signal and the normal signal data are spliced in the original order, a point-by-point analysis algorithm is used to detect the data characteristics of the correction splicing boundary, the boundary data characteristic detection is completed by volatility rate and mean difference analysis, the volatility rate detection is completed by point-by-point calculation of variance and comparison with the variance of the neighborhood data, the mean difference is completed by calculating the mean difference between the boundary point and its neighborhood point, the connection value is adjusted by using a data consistency adjustment method, the adjustment method is a moving average method, and the window size is set to 5 sampling points, the overall volatility rate of the spliced signal is globally detected, the deviation of the cross-segment value is corrected, and the reconstructed signal data is generated;

[0082] The reconstructed signal data includes a reconstructed waveform, an adjusted frequency response and a signal stability score.

[0083] Please refer to Figure 4 The specific steps of generating the abnormal mode and cause correlation report are:

[0084] S301: Based on the reconstructed signal data, the signal data is divided into multiple independent hierarchical data regions according to the time range and signal amplitude change trend, the change rate and structural features of the signal are extracted and processed by layer-by-layer scanning, the key parameters of the signal are extracted, and the abnormal data part existing in the signal is separated, the dynamic characteristic separation of the abnormal data is completed, and signal decomposition feature data is generated;

[0085] S302: Based on the signal decomposition feature data, the generator of the generative adversarial network learns the latent distribution characteristics of the signal decomposition feature data, compares the generated data and the actual data, captures the abnormal mode features in the signal distribution, extracts the abnormal mode features as key dynamic parameters, matches the abnormal mode features with the known characteristics of potential abnormal causes, completes the correlation mapping of the mode and the cause, and arranges the analysis results into a multi-dimensional parameter set to generate abnormal mode and cause relationship data;

[0086] S303: Based on the abnormal pattern and cause relationship data, the causal characteristics of the abnormal pattern are modeled using a generative adversarial network, the causal characteristic distribution is reconstructed, and the consistency of the generated distribution and the actual distribution is evaluated to determine the dynamic correlation path of the causal variables. The variables in the causal path are analyzed item by item using a feature relationship mapping method, the causal strength between the pattern characteristics and the abnormal causes is quantified, the modeling results are recorded as causal correlation data sets, and a causal relationship model is obtained. At the same time, an abnormal pattern and causal correlation report is generated;

[0087] Based on the reconstructed signal data, the signal data is partitioned according to the time range and signal amplitude change trend using a hierarchical partitioning algorithm. The change trend of the signal is extracted by setting a sliding window with a time window size of 100 milliseconds. Each partition is aggregated according to the amplitude change direction. The change rate of each segment of the signal is quantified using a change rate calculation formula. The quantization method is to divide the amplitude difference in each time window by the time difference. The frequency characteristics of the signal are extracted using the fast Fourier transform method. The main frequency value and high frequency component are extracted after converting the amplitude sequence to the frequency domain signal. The absolute amplitude of the signal change is extracted using the amplitude difference calculation method. The amplitude calculation formula is the sum of the absolute values of the amplitude difference between adjacent sampling points. The abnormal data is separated from the normal signal by comparing the signal strength. The dynamic change characteristics of the abnormal data are recorded, and signal decomposition feature data is generated.

[0088] Based on the signal decomposition feature data, the generator of the generative adversarial network learns the latent distribution characteristics of the signal decomposition feature data. The generator uses a deep convolutional neural network structure with a convolution kernel size of 3x3 and a channel number of 64. After generating a contrast sample of feature data, the discriminator discriminates the distribution difference between the generated sample and the actual sample. The discriminator uses a binary classification structure and a cross-entropy loss function. The feature comparison algorithm is used to capture the abnormal pattern characteristics existing in the distribution of the generated data and the actual data. The feature extraction function is used to extract the abnormal pattern characteristics as key dynamic parameters, including signal amplitude range, frequency offset, and jump point position. The pattern matching method is used to match the abnormal pattern characteristics with the known characteristics of potential abnormal causes. The association matrix is used to construct the association mapping relationship between the pattern and the cause. The association results between the pattern and the cause are stored as a multi-dimensional parameter set, and abnormal pattern and cause relationship data are generated.

[0089] Based on the abnormal pattern and cause relationship data, the causal characteristics of the abnormal pattern are modeled using a generative adversarial network. The generator uses a bidirectional long short-term memory network structure to capture the signal causal distribution. The input layer node number is set to 256, the hidden layer node number is set to 128, and the time step is set to 10 steps. The mean square error is used as the loss function to optimize the generated distribution. After reconstructing the causal characteristic distribution, the consistency of the generated distribution and the actual distribution is calculated to determine the dynamic correlation path of the causal variable. The variables in the causal path are analyzed one by one using a feature relationship mapping method. The causal strength between the pattern characteristics and the abnormal causes is quantified using a weight normalization algorithm. The weight normalization method is to divide the weight values of all causal paths by the sum of the total weight values. The causal correlation data set of the modeling result is generated, the distribution characteristics of the causal variables are recorded, and the causal relationship model is obtained. At the same time, an abnormal pattern and causal correlation report is generated.

[0090] The abnormal pattern and causal correlation report includes a list of identified abnormal patterns, related cause analysis, and impact degree evaluation.

[0091] The data structure is processed using a multi-level feature decomposition method to extract feature parameters from each layer of data. The signal data is initially layered according to time period and frequency range. The data is divided into multiple different characteristic intervals. The amplitude, rate of change, and structure parameters of the layered data are extracted through layer-by-layer scanning. The dynamic characteristics of each layer of data are grouped and aggregated. The change range and trend of the fluctuation amplitude are extracted. The extracted dynamic characteristics are compared with the abnormal parameters layer by layer. The change range of the abnormal characteristics is recorded. The abnormal signal content with dynamic characteristics is separated.

[0092] The pattern characteristics in the abnormal signal are identified. The pattern characteristics are matched with potential abnormal causes. The dynamic characteristics related to time series changes in the data are extracted. The amplitude difference, periodic change, and signal mutation point in different time points and frequency ranges are compared. The characteristic distribution of the data is filtered one by one. The abnormal signal characteristic range is found. The fluctuation pattern is further analyzed. The repeated change period, peak interval, and change rule are extracted. The potential abnormal cause change pattern is matched with the existing characteristic parameter set. The specific association information between the pattern and the cause is recorded.

[0093] Please refer to Figure 5 The specific steps for generating the error source quantitative analysis result are as follows:

[0094] S401: Based on the abnormal pattern and causal correlation report, the causal factor information in the report is extracted. The weight of the multivariate relationship is analyzed. The correlation strength and weight parameters between the multivariables are decomposed. The causal influence weight distribution result is generated.

[0095] S402: Based on the causal influence weight distribution result, extract error data in the variable, perform statistical analysis of error bias, evaluate error sources through parameter difference decomposition, and generate error source statistical result;

[0096] S403: Based on the error source statistical result, perform quantitative decomposition of error sources, extract error distribution characteristics of causal variables, and complete quantitative evaluation by combining error weight calculation, to generate error source quantitative analysis result;

[0097] Based on the abnormal pattern and causal correlation report, extract the causal factor information in the report, use multivariate weight analysis method to analyze the relationship between causal variables, use weight decomposition algorithm to calculate the correlation strength between multiple variables, the weight decomposition process includes constructing a variable relationship matrix, the matrix dimension is equal to the number of variables, calculating the mutual information value between each pair of variables and normalizing it to correlation weight value, using matrix decomposition method to decompose the weight parameter, the decomposition method is singular value decomposition, extract the feature vector in the decomposition result and store it as a weight parameter table, and generate the causal influence weight distribution result;

[0098] Based on the causal influence weight distribution result, extract error data in the variable, use bias statistical analysis method to analyze and process error data, and analyze error sources through difference decomposition method, the difference decomposition method is to calculate the absolute error of the observed value and the theoretical value of the variable, the absolute error is defined as the absolute value of the observed value minus the theoretical value, use distribution analysis algorithm to perform distribution statistics on error data, the distribution statistics method is to construct an error histogram, the number of intervals of the histogram is set to 10, the interval width is the total error range divided by the number of intervals, save the statistical result as an error distribution table, and generate an error source statistical result;

[0099] Based on the error source statistical result, extract the error distribution characteristics of causal variables, use quantitative decomposition method to quantitatively analyze error sources, the quantitative decomposition method is to combine the weight value in the error distribution table to perform weighted decomposition on error data, the weighted method is to multiply each error value by its corresponding weight value, use normalization method to convert error result to percentage form, mark key items in error distribution characteristics, the marking method is to select items with error percentage greater than 10, combine error weight calculation to complete quantitative evaluation of error sources, store all evaluation data as an error quantization table, and generate an error source quantitative analysis result;

[0100] The error source quantitative analysis result includes a main error source list, the contribution rate of each source, and the environmental factors affecting the error.

[0101] Please refer to Figure 6 The specific steps of generating the reverse corrected data are as follows:

[0102] S501: Based on the error source quantitative analysis result, extract the specific parameters in the error distribution, group analysis each parameter according to the deviation range, calculate the deviation value of each group parameter, adjust the data range in the group, and generate a set of corrected deviation parameters;

[0103] S502: Based on the set of corrected deviation parameters, correct each parameter in the original data one by one, replace the data with the corrected value in the grouping range, correct the deviation value by cumulative method, and generate the corrected original data;

[0104] S503: Based on the corrected original data, extract the residual abnormal values in the data that have not been corrected, evaluate the out-of-range deviation of the abnormal values, gradually adjust the deviation parameters of the abnormal values, and generate the data after reverse correction;

[0105] Based on the error source quantitative analysis result, extract the specific parameters in the error distribution, use the deviation grouping algorithm to group analyze each parameter according to the deviation range, the grouping method is to set the grouping interval, the interval range is 10 equal parts from the minimum value to the maximum value of the deviation value, calculate the deviation value of each group parameter using the group deviation mean formula, the formula is the weighted average of all deviation values in the group, the weight is the influence factor of the parameter, adjust the upper and lower limits of the data in the group using the range adjustment algorithm, the adjustment rule is to expand the upper and lower limits by 10% of the deviation between the maximum and minimum values in the group, store all adjusted data as a parameter set, and generate a set of corrected deviation parameters;

[0106] Based on the set of corrected deviation parameters, correct each parameter in the original data one by one, replace the data with the corrected value in the grouping range using the range correction method, the correction value is defined as the group deviation mean, locate the corresponding parameter position in the original data through the matching algorithm, replace each matching item with the correction value, and correct the deviation value by cumulative method, the cumulative method is to add the current deviation value and the previous correction result and update the current parameter value, the cumulative result is stored in the corrected parameter array, and the corrected original data is generated;

[0107] Based on the corrected original data, extract the residual abnormal values in the data that have not been corrected, evaluate the out-of-range deviation of the residual abnormal values using the abnormal value detection algorithm, the range of abnormal values is calculated by setting the upper and lower limits, the upper limit is the 95% quantile value of normal data, and the lower limit is the 5% quantile value, values outside the range are marked as abnormal values, gradually adjust the deviation parameters of the abnormal values, the adjustment method is to decrease the difference between the abnormal value and its adjacent normal value by 10%, until the abnormal value returns to the range, store the adjusted abnormal values as a set of reverse correction data, and generate the data after reverse correction;

[0108] The inverse correction data includes adjusted data parameters, data comparison before and after optimization, and correction effect evaluation.

[0109] Please refer to Figure 7 The specific steps for generating the metrology data category recognition result are as follows:

[0110] S601: Based on the inverse correction data, scan the classification feature values in the data item by item, extract the feature range of each data item, add a mark value to the data according to the classification standard, and generate a data classification feature set;

[0111] S602: Based on the data classification feature set, use a support vector machine to compare and analyze the inverse correction data and the classification feature values, gradually confirm the classification mark of each data item, classify and organize the data according to the mark result, and generate a data classification mark result;

[0112] S603: Based on the data classification mark result, input each classification mark into a deep learning model, perform hierarchical organization of the classification output combined with the input data, complete the extraction of the classified metrology data, and generate a metrology data category recognition result;

[0113] Based on the inverse correction data, scan the classification feature values in the data item by item, use an interval extraction algorithm to calculate the feature range of each data item, the feature range is defined as the difference between the minimum and maximum values of the feature value, read the feature value sequence one by one through the traversal algorithm, and count the maximum and minimum values of each feature value. Using the set classification standard, match and analyze the feature value and the standard value, the matching method is to compare the range of the feature value with the upper and lower limits of the classification standard, add the corresponding mark value to the feature data table, the mark value is in the format of integer data, and is stored in the classification feature matrix to generate a data classification feature set;

[0114] Based on the data classification feature set, use a support vector machine to compare and analyze the inverse correction data and the classification feature values, the support vector machine model uses a radial basis function kernel, the kernel parameter γ is set to 0.1, and the penalty factor C is set to 1. Train the support vector machine model using the training set data, input the feature value vector in the classification feature set into the model, and classify and predict each data item. The prediction category output is an integer data, gradually confirm the classification mark of each data item, classify and organize the data according to the classification mark through a grouping algorithm, the grouping method is to store according to the classification mark value, save the grouped and organized data to a classification result table, and generate a data classification mark result;

[0115] Based on the data classification label result, each classification label is input to a deep learning model, the deep learning model adopts a full connection neural network, the network structure includes an input layer, two hidden layers and an output layer, the number of nodes of the input layer is the dimension of the classification label, the number of nodes of the hidden layers is 128 and 64 respectively, the number of nodes of the output layer is the number of categories of the classification output, the loss function of the model is cross entropy loss, the learning rate is set to 0.001, the classification label and the original data are combined as an input feature vector which is input to the deep learning model, the feature vector is classified and predicted by the model, the classification result is stored in the hierarchical data table according to the category, the classification result data is layered and grouped according to the hierarchical arrangement algorithm, the layered result is saved to the classification result data set, and the measurement data category recognition result is generated;

[0116] Among them, the measurement data category recognition result includes the data type of classification, the accuracy of classification and the decision basis of model judgment.

[0117] Support vector machine, according to the formula:

[0118]

[0119] Among them: f'(x) is the classification label result of measurement detection data, x is the feature vector of the measurement detection data to be identified, x i is the feature vector of the support vector which has been trained in the support vector machine, α i is the Lagrange multiplier weight of the support vector, y i is the category label corresponding to the support vector, exp(-γ||x i -x|| 2 ) is the Gaussian kernel function, γ is the scale parameter of the Gaussian kernel, θ is the dynamic adjustment parameter, w is the feature weighting vector, is the feature weighting result of the linear part of the kernel function, b is the bias term of the model, λ||x|| 2 is the regularization term, λ is the regularization coefficient;

[0120] Execution process: first, the support vector machine is trained by the training set, and the feature vector x i of the support vector, the corresponding category label y i , the Lagrange multiplier weight α i and the model bias term b are obtained, then for the measurement detection data x to be identified, the similarity of the to-be-identified data and the support vector in the feature space is calculated using the kernel function, wherein the nonlinear part is realized by the Gaussian kernel, the kernel scale parameter γ is obtained by cross validation method, and the linear part The importance of specific features is emphasized by a feature weighting vector w and a dynamic adjustment parameter θ, w is calculated according to the information gain of each feature in the metrology detection task, θ is dynamically optimized by gradient descent method to adjust the weight ratio of nonlinear and linear parts, and a regularization term λ||x|| is added 2 The influence of noise on the result is inhibited, the regularization coefficient λ is determined by the grid search method, and is used to balance the complexity and generalization ability of the model, the kernel function results of all support vectors are accumulated and added to the bias term b and the regularization term λ||x|| 2 The calculation result is mapped to a classification label, and the classification label result of the metrology detection data is output.

[0121] The classification feature values in the data are scanned item by item, and the feature range of each item of data is extracted, the classification feature values are attributes and indexes used for classifying in the data, and the extraction step specifically comprises the following steps: from the data after reverse correction, numerical type and classification type attributes are extracted according to a predefined classification rule, for numerical type attributes, numerical values are directly extracted, for classification type attributes, a dictionary mapping is used to convert them into numerical codes, for missing values, interpolation and default values are used for filling, finally, the feature values are sorted and classified.

[0122] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms, any skilled person in the art can modify or change the above disclosed technical content to equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application to the above embodiments without departing from the technical solution content of the present application still belongs to the protection scope of the present application technical solution.

Claims

1. A method for automatic identification of metrological testing data based on deep learning, characterized in that, Includes the following steps: Step 1: Receive raw data through the sensor, and simultaneously mark and record distorted sections by scoring the signal strength, generating a signal quality assessment report; Step 2: Based on the signal quality assessment report, wavelet transform is used to correct the marked segments and the distorted signal to obtain the reconstructed signal data. Step 3: Perform structural decomposition on the reconstructed signal data, identify abnormal patterns and causes through generative adversarial networks, continuously train the generative adversarial network to obtain a deep learning model, record the relationship between abnormal patterns and causes, and generate an abnormal pattern and causal association report. Step 4: Using the aforementioned anomaly pattern and causal association report, calculate the influence intensity of each factor through multiple regression analysis, and simultaneously assess the sources of error to obtain quantitative analysis results of the error sources; Step 5: Based on the quantitative analysis results of the error sources, the original data is corrected by precisely adjusting the parameters to optimize the accuracy and reliability of the original data and generate the reverse-corrected data; Step Six: Based on the reverse-corrected data, extract the feature range of each data item, add label values ​​to the data according to the classification criteria, and use a support vector machine to compare and analyze the reverse-corrected data with the classification feature values, output the classification results, and input each classification label into the deep learning model to obtain the quantitative data category recognition results.

2. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The signal quality assessment report includes the specific location of the distorted section, the signal strength level, and the noise level.

3. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The reconstructed signal data includes the reconstructed waveform, the adjusted frequency response, and the signal stability score.

4. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The specific steps for generating the aforementioned anomaly pattern and causal relationship report are as follows: Based on the reconstructed signal data, the signal data is divided into multiple independent hierarchical data regions according to the time range and the trend of signal amplitude change. The rate of change and structural features of the signal are extracted by scanning layer by layer, the key parameters of the signal are extracted, and the abnormal data parts in the signal are separated to complete the dynamic characteristic separation of abnormal data and generate signal decomposition feature data. Based on the signal decomposition feature data, the generator of the generative adversarial network is used to learn the potential distribution characteristics of the signal decomposition feature data. By comparing the generated data with the actual data, abnormal pattern features in the signal distribution are captured, and the abnormal pattern features are extracted as key dynamic parameters. The abnormal pattern features are matched with the known characteristics of potential abnormal causes to complete the association mapping between patterns and causes. The analysis results are organized into a multi-dimensional parameter set to generate abnormal pattern and cause relationship data. Based on the aforementioned abnormal pattern and causal relationship data, generative adversarial networks are used to model the causal characteristics of abnormal patterns. At the same time, the causal characteristic distribution is reconstructed, and the consistency between the generated distribution and the actual distribution is evaluated. The dynamic association path of causal variables is determined, and the variable in the causal path is analyzed item by item using the feature relationship mapping method. The causal strength between pattern features and abnormal causes is quantified, and the modeling results are recorded as a causal association dataset to obtain a causal relationship model. At the same time, an abnormal pattern and causal association report are generated. The abnormal pattern and causal association report includes a list of identified abnormal patterns, related causal analysis, and an assessment of the degree of impact.

5. The method for automatic identification of metrological testing data based on deep learning according to claim 4, characterized in that, The data structure is processed by a multi-level feature decomposition method, and feature parameters are extracted from each layer of data. The signal data is initially layered according to time period and frequency range, and the data is divided into multiple different characteristic intervals. The amplitude, rate of change and structural parameters are extracted for the layered data by scanning layer by layer. The dynamic characteristics of each layer of data are grouped and aggregated. At the same time, the range and trend of fluctuation amplitude are extracted. The decomposed dynamic features are compared with the abnormal parameters layer by layer, the range of abnormal characteristics is recorded, and the abnormal signal content with dynamic characteristics is separated. Identify pattern features in abnormal signals, match pattern features with potential causes of anomalies, extract dynamic characteristics related to time series changes in the data, and filter the characteristic distribution of the data item by item by comparing amplitude differences, periodic changes and signal mutation points in different time points and frequency ranges to find the range of abnormal signal features. Further analyze the fluctuation patterns, extract the recurring change period, peak spacing and change patterns, and match them to the existing set of characteristic parameters. Compare the change patterns of potential causes of anomalies and record the specific correlation information between patterns and causes.

6. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The quantitative analysis results of the error sources include a list of major error sources, the contribution rate of each source, and environmental factors affecting the error.

7. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The reverse-corrected data includes adjusted data parameters, a comparison of data before and after optimization, and an evaluation of the correction effect.

8. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The specific steps for generating the measurement data category identification results are as follows: Based on the reverse-corrected data, the classification feature values ​​in the data are scanned item by item, the feature range of each data item is extracted, and a label value is added to the data according to the classification standard to generate a data classification feature set; Based on the data classification feature set, a support vector machine is used to compare and analyze the reverse-corrected data with the classification feature values, gradually confirm the classification label of each data item, and classify and organize the data based on the labeling results to generate data classification labeling results. Based on the data classification and labeling results, each classification label is input into the deep learning model, and the classification output is hierarchically organized in combination with the input data to complete the extraction of the classified measurement data and generate measurement data category recognition results. The measurement data category identification results include the data type of the classification, the accuracy of the classification, and the decision basis for the model judgment.

9. The method for automatic identification of metrological testing data based on deep learning according to claim 1, characterized in that, The support vector machine is configured according to the formula: ， in: The classification and labeling results of measurement and testing data, The feature vector of the measurement and testing data to be identified. These are the feature vectors of the support vectors that have been trained in the support vector machine. For the Lagrange multiplier weights of the support vectors, For the category labels corresponding to the support vectors, For Gaussian kernel function, Let be the scale parameter of the Gaussian kernel. To dynamically adjust parameters, For feature weighting vectors, The weighted result of the features of the linear part of the kernel function. For the bias term of the model, For regularization terms, is the regularization coefficient.

10. The method for automatic identification of metrological testing data based on deep learning according to claim 8, characterized in that, The categorical feature values ​​in the data are scanned item by item to extract the feature range of each data item. Categorical feature values ​​are the attributes and indicators used to distinguish categories in the data. The extraction steps are as follows: from the reverse-corrected data, numerical and categorical attributes are extracted according to predefined classification rules. For numerical attributes, the values ​​are extracted directly. For categorical attributes, dictionary mapping is used to convert them into numerical codes. For missing values, interpolation and default values ​​are used to fill them. Finally, the feature values ​​are sorted and classified.

Images