Instrument metering detection and calibration method and system

Abstract

The invention relates to the technical field of detection and calibration, in particular to an instrument metering detection and calibration method and system. The method comprises the following steps: analyzing a running state log of a current meter, extracting multi-frequency instantaneous overload current waveform time domain data, and carrying out shunting resistor resistance drift simulation analysis based on the data to obtain resistance drift data; then, quantizing the drift data to obtain a resistance drift intensity segmentation index, and based on the resistance drift intensity segmentation index, carrying out pre-amplifier gain error nonlinear quantization to obtain gain error data; thirdly, error increment change data is obtained through error change multi-point calibration, multi-point segmentation error calibration is carried out based on the error increment change data, and finally multi-point segmentation error calibration learning data is obtained. According to the method, the detection and calibration technology is optimized, so that the detection and calibration technology is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of detection and calibration, and particularly to a method and system for instrument metrological detection and calibration. Background Art

[0002] The accuracy of current meters is directly related to the stability and safety of the system, especially in application scenarios with high-precision requirements, such as the fields of electricity, automation control, communication, etc. However, during the long-term use of current meters, due to the influence of factors such as temperature changes, aging effects, and mechanical vibrations, the performance of key components inside them (such as shunt resistors and preamplifiers) will change to a certain extent, resulting in measurement errors. These errors are manifested as gain errors, resistance value drifts, etc., which in turn affect the measurement accuracy of current meters. In particular, the resistance value drift of shunt resistors and the gain error of preamplifiers gradually intensify over time and become the main factors affecting the accuracy of current meters. Such drift not only affects the immediate measurement data of the meter but also leads to long-term error accumulation in the entire measurement system, thereby affecting the final monitoring results. Therefore, how to detect and calibrate these errors in a timely and accurate manner has become a key issue in improving the performance of current meters. However, there is a problem in the traditional method for instrument metrological detection and calibration that the analysis of the resistance value drift of shunt resistors and the gain error of preamplifiers is inaccurate, resulting in large errors in the metrological detection and calibration of current meters. Summary of the Invention

[0003] Based on this, it is necessary to provide a method and system for instrument metrological detection and calibration to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for instrument metrological detection and calibration, the method includes the following steps: Step S1: Obtain the operation status log of the current meter; extract the multi-frequency instantaneous overload current from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain the multi-frequency instantaneous overload current waveform time-domain data; Step S2: Based on the multi-frequency instantaneous overload current waveform time-domain data, perform simulation analysis on the resistance value drift of the shunt resistor to obtain the shunt resistor resistance value drift data; perform piecewise exponential quantization on the shunt resistor resistance value drift data for the drift intensity to obtain the piecewise exponential of the resistance value drift intensity; perform non-linear quantization on the gain error of the preamplifier according to the piecewise exponential of the resistance value drift intensity to obtain the preamplifier gain error quantization data; Step S3: Perform multi-point calibration of the error change based on the preamplifier gain error quantization data to obtain the multi-point calibration data of the error increment change; perform multi-point piecewise error calibration based on the multi-point calibration data of the error increment change to obtain the multi-point piecewise error calibration learning data.

[0005] Preferably, step S1 includes the following steps: Step S11: Obtain the operation status log of the current meter; Step S12: Clean the operation status log of the current meter to obtain the cleaned operation status log of the current meter; Step S13: Extract multi-frequency instantaneous overload currents from the cleaned operation status log of the current meter to obtain multi-frequency instantaneous overload current data; Step S14: Perform waveform time-domain analysis on the multi-frequency instantaneous overload current data to obtain multi-frequency instantaneous overload current waveform time-domain data.

[0006] Preferably, step S2 includes the following steps: Step S21: Calculate the overshoot repeat ratio difference of the overload waveform based on the multi-frequency instantaneous overload current waveform time-domain data to obtain the overshoot repeat ratio difference of the overload current; Step S22: Perform simulation analysis of the shunt resistance value drift according to the overshoot repeat ratio difference of the overload current to obtain the shunt resistance value drift data; Step S23: Perform drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain the segmented exponential of the resistance value drift intensity; Step S24: Perform non-linear quantization of the gain error of the preamplifier according to the segmented exponential of the resistance value drift intensity to obtain the gain error quantization data of the preamplifier.

[0007] Preferably, step S23 includes the following steps: Step S231: Analyze the relative change amount of the shunt resistance value drift of the shunt resistance value drift data to obtain the relative change amount of the shunt resistance value drift; Step S232: Identify the multi-stage logarithmic gradient of the increasing shunt resistance value according to the relative change amount of the shunt resistance value drift of the shunt resistance value drift data to obtain the multi-stage logarithmic gradient of the increasing shunt resistance value; Step S233: Calculate the growth difference of the discrete monotonic increasing rate for the multi-stage logarithmic gradient of the increasing shunt resistance value to obtain the growth difference of the discrete monotonic increasing rate; Step S234: Fit the extreme value interval of the drift segmented intensity for the stage logarithmic gradient of the increasing shunt resistance value based on the growth difference of the discrete monotonic increasing rate to obtain the fitting data of the shunt resistance value drift segmented interval intensity; Step S235: Perform drift intensity segmented exponential quantization according to the fitting data of the shunt resistance value drift segmented interval intensity to obtain the segmented exponential of the resistance value drift intensity.

[0008] Preferably, step S24 includes the following steps: Step S241: Perform multi-scale difference processing on the segmented exponential of the resistance value drift intensity to obtain the multi-scale difference data of the resistance value drift intensity; Step S242: Based on the multi-scale differential data of the resistance value drift intensity, perform distortion simulation on the output voltage signal of the preamplifier to obtain the distortion data of the output voltage signal; Step S243: Analyze the non-linear amplification disorder of the output voltage signal distortion data to obtain the non-linear amplification disorder data of the voltage signal; Step S244: Perform time-series dependent signal interpolation processing according to the non-linear amplification disorder data of the voltage signal to obtain the disorder interpolation data of the voltage signal; Step S245: Perform non-linear quantization of the gain error of the preamplifier according to the disorder interpolation data of the voltage signal to obtain the quantization data of the gain error of the preamplifier.

[0009] Preferably, Step S244 includes the following steps: Calculate the average difference of the signal change amplitudes before and after the time series for the non-linear amplification disorder data of the voltage signal to obtain the average difference of the signal change amplitudes in the time series; Perform signal change variational inference processing based on the average difference of the signal change amplitudes in the time series to obtain approximate data of the signal change distribution difference; Perform marginal likelihood estimation on the approximate data of the signal change distribution difference to obtain the marginal likelihood estimation data of the change difference; Perform time-series dependent signal interpolation processing on the marginal likelihood estimation data of the change difference based on the Bayesian interpolation method to obtain the disorder interpolation data of the voltage signal.

[0010] Preferably, Step S3 includes the following steps: Step S31: Analyze the trend of the time-series error increment for the quantization data of the gain error of the preamplifier to obtain the trend data of the time-series error increment; Step S32: Perform multi-point calibration of the error change according to the trend data of the time-series error increment to obtain the multi-point calibration data of the error increment change; Step S33: Perform multi-point segmented error calibration based on the multi-point calibration data of the error increment change to obtain the multi-point segmented error calibration data; Step S34: Perform logical learning on the multi-point segmented error calibration data to obtain the multi-point segmented error calibration learning data.

[0011] Preferably, Step S33 includes the following steps: Step S331: Perform error mutation difference operation on the multi-point calibration data of the error increment change to obtain the multi-point mutation difference data of the error; Step S332: Perform multiple regression analysis on the multi-point mutation difference data of the error to obtain the multi-point mutation regression data of the error; Step S333: Calculate the non-constant variance of the multi-point mutation of the error based on the multi-point mutation regression data of the error to obtain the non-constant variance of the multi-point mutation of the error; Step S334: Perform time-series multi-point error structure similarity analysis on the error multi-point mutation regression data to obtain time-series multi-point error structure similar data; Step S335: Set the numerical change of error equal compensation according to the error multi-point mutation non-constant variance and the time-series multi-point error structure similar data to obtain the error equal compensation numerical setting data; Step S336: Perform multi-point segmented error calibration based on the error equal compensation numerical setting data to obtain multi-point segmented error calibration data.

[0012] Preferably, the present invention also provides an instrument measurement detection and calibration system for executing the instrument measurement detection and calibration method as described above. The instrument measurement detection and calibration system includes: A waveform time-domain analysis module, configured to obtain the operation status log of the current meter; extract the multi-frequency instantaneous overload current from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain the multi-frequency instantaneous overload current waveform time-domain data; An error non-linear quantization module, configured to perform shunt resistance value drift simulation analysis based on the multi-frequency instantaneous overload current waveform time-domain data to obtain shunt resistance value drift data; perform drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain the resistance value drift intensity segmented exponent; perform non-linear quantization of the gain error of the preamplifier according to the resistance value drift intensity segmented exponent to obtain the preamplifier gain error quantization data; A multi-point segmented error calibration module, configured to perform multi-point calibration of error changes according to the preamplifier gain error quantization data to obtain multi-point calibration data of error increment changes; perform multi-point segmented error calibration based on the multi-point calibration data of error increment changes to obtain multi-point segmented error calibration learning data.

[0013] The beneficial effects of the present invention are as follows. By obtaining the operation status log of the current meter and extracting the multi-frequency instantaneous overload current from it, the working status of the current meter under different load conditions can be accurately captured. This process can help identify the response characteristics of the current meter under instantaneous overload conditions, especially its behavior under high load or extreme working environments, avoiding the dynamic errors that cannot be fully reflected by traditional static measurement methods. Through waveform time-domain analysis, the time-domain data of the multi-frequency instantaneous overload current waveform is obtained, providing reliable raw data support for subsequent error analysis and calibration. Based on the time-domain data of the multi-frequency instantaneous overload current waveform, the simulation analysis of the shunt resistance value drift can be carried out to simulate the resistance value drift characteristics of the current meter in the actual working environment, and clearly identify the trend of the shunt resistance value changing with time and working conditions. Through the piecewise exponential quantization of the drift intensity, different degrees of resistance value drift can be quantified, so as to more accurately evaluate the performance change of the current meter. Then, based on the piecewise exponential of the resistance value drift intensity, the gain error of the preamplifier is non-linearly quantified, which can effectively predict the change of the gain error and provide a scientific basis for subsequent error calibration. This process can greatly improve the accuracy and stability of the current meter. According to the quantization data of the preamplifier gain error, the multi-point calibration of the error change is carried out, which can accurately calibrate the error at multiple working points and help identify the change of the gain error under different working conditions. This multi-point calibration method is more comprehensive and accurate than the traditional single-point calibration method and can effectively cover various states during the operation of the instrument. Subsequently, based on the multi-point piecewise error calibration of the error increment change data, targeted calibration can be carried out for different error intervals, further improving the measurement accuracy of the current meter. Through this multi-point piecewise error calibration method, the measurement deviation caused by the shunt resistance drift and the gain error can be significantly reduced, and the reliability and accuracy of the current meter in practical applications can be improved. Therefore, the present invention is an optimized treatment of a traditional instrument measurement, detection and calibration method, solving the problem that the traditional instrument measurement, detection and calibration method has inaccurate analysis of the shunt resistance value drift and the preamplifier gain error, resulting in large errors in the instrument measurement, detection and calibration, improving the accuracy of the analysis of the shunt resistance value drift and the preamplifier gain error, and reducing the error of the instrument measurement, detection and calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic flow chart of the steps of an instrument measurement, detection and calibration method; Figure 2 is Figure 1 a detailed implementation step flow chart of step S2 in Figure 3 is Figure 1 a detailed implementation step flow chart of step S3 in DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Please refer to Figures 1 to 3 , an instrument measurement detection and calibration method, the method comprising the following steps: Step S1: Obtain the operation status log of the current meter; extract multi-frequency instantaneous overload currents from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain multi-frequency instantaneous overload current waveform time-domain data; Step S2: Based on the multi-frequency instantaneous overload current waveform time-domain data, perform shunt resistance value drift simulation analysis to obtain shunt resistance value drift data; perform drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain a segmented exponential of the resistance drift intensity; perform gain error non-linear quantization of the preamplifier according to the segmented exponential of the resistance drift intensity to obtain preamplifier gain error quantization data; Step S3: Perform error change multi-point calibration according to the preamplifier gain error quantization data to obtain error increment change multi-point calibration data; perform multi-point segmented error calibration based on the error increment change multi-point calibration data to obtain multi-point segmented error calibration learning data.

[0016] In an embodiment of the present invention, referring to Figure 1 as described, it is a schematic diagram of the step flow of an instrument measurement detection and calibration method of the present invention. In this example, the instrument measurement detection and calibration method comprises the following steps: Step S1: Obtain the operation status log of the current meter; extract multi-frequency instantaneous overload currents from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain multi-frequency instantaneous overload current waveform time-domain data; In an embodiment of the present invention, by collecting the operation status log generated during the continuous operation of a specific model (for example, with a rated measurement range of 0-100A, a sampling frequency of 100kHz, and a storage capacity of 32MB) of the current meter, the log includes but is not limited to the real-time sampling value of the current channel, the sampling timestamp, the voltage power supply status, the fault flag code, and the environmental temperature record. Subsequently, the log is screened and denoised in multiple rounds using regular matching rules and a feature event timestamp screening mechanism, and records in non-measurement working states such as device startup self-check and no-load operation are excluded. By setting a current mutation threshold (such as setting it above 150% of the rated value, that is, greater than 150A), multi-frequency instantaneous overload current events are extracted. The 10ms sampling windows before and after each sudden overload current event are locked on the time axis, and the current waveform data within these windows are used as the analysis input. The joint algorithm of fast Fourier transform (FFT) and wavelet packet transform (WPT) is used to reconstruct the multi-scale time-domain features of the extracted current waveform data. By calculating typical indicators such as the rising edge rate, peak duration, and falling edge steepness of each event waveform, a complete waveform time-domain parameter data sequence of each instantaneous overload event is finally output, and multi-event alignment is performed on a unified time axis.

[0017] Step S2: Based on the time-domain data of the multi-frequency instantaneous overload current waveform, perform a shunt resistance value drift simulation analysis to obtain shunt resistance value drift data; perform a piecewise exponential quantization of the shunt resistance value drift data to obtain a piecewise exponential of the resistance value drift intensity; perform a non-linear quantization of the gain error of the preamplifier according to the piecewise exponential of the resistance value drift intensity to obtain preamplifier gain error quantization data; In the embodiment of the present invention, a shunt resistance value drift simulation analysis is performed based on the extracted waveform time-domain parameter data. First, 10 representative types of typical overload waveforms are selected. Through multi-dimensional clustering analysis (using the K-means algorithm with the number of clusters set to 10) of indicators such as the peak current, duration, and rising edge slope of each waveform, each type of waveform is used as a loading template and correspondingly input into the thermal-electrical coupling finite element simulation framework. In this framework, the material parameters of the shunt resistance are set (for example, manganin is selected with a temperature coefficient of +0.00002 / °C), and the current density time series corresponding to the clustered waveform is applied. By combining the transient heat conduction equation with the heat generation power density of the resistance power consumption, the internal temperature gradient is calculated, and then the dynamic change of the resistance value at different time nodes is solved. The offset of the resistance value relative to the initial resistance value after each loading is recorded and output as the resistance value drift data. Perform a sliding window process on this data, set the window length to 100 events, and the step size to 10. Calculate the drift intensity by dividing the difference between the maximum offset and the minimum offset of the resistance value within the window by the initial resistance value. Perform a piecewise process on the intensity values of all windows (set the interval thresholds to 0.5%, 1%, 1.5%, 2%), and map the intensity intervals to fixed exponential labels (1, 2, 3, 4) respectively to complete the piecewise exponential quantization of the shunt resistance value drift intensity. After obtaining this exponent, further construct a gain error mapping table. Measure and verify the output errors of the preamplifier corresponding to each exponent level at gain configuration values of 1 times, 10 times, and 100 times. Through non-linear fitting (such as using cubic spline interpolation) of the residual values between the output errors and the gain multiples, obtain the non-linear quantization values of the gain errors corresponding to each intensity segment to form a gain error quantization data table.

[0018] Step S3: Perform multi-point calibration of the error change according to the preamplifier gain error quantization data to obtain multi-point calibration data of the error increment change; perform multi-point segmented error calibration based on the multi-point calibration data of the error increment change to obtain multi-point segmented error calibration learning data.

[0019] In the embodiments of the present invention, the obtained preamplifier gain error quantization data is sorted in chronological order, and a sliding window error change trend analysis process is performed throughout the time period. The window length is set to 1000 data points, and the step size is 200 data points. The linear change rate and change direction of the gain error amount within each window are calculated and recorded as the time series error increment trend data. Then, based on the trend data, 10 time periods with the most drastic error change amplitudes are extracted, and distribution sampling is performed on the error data points within these time periods. 10 representative points are selected for each period and recorded as the error increment change multi-point calibration data. Next, an error calibration operation is performed on these data. A piecewise linear approximation calibration method is adopted to construct a least squares linear correction model within each error interval (i.e., perform linear regression on the gain error and the actual deviation value), and the corresponding correction coefficients are respectively fitted for multiple intervals. The correction coefficient combinations of all intervals form the multi-point piecewise error calibration data. Finally, a logical relationship structure is further introduced into these calibration models. By analyzing the differences in the calibration correction coefficients between the models in different intervals, the error interval division logic is made logical using a tree discriminant structure, and a logical function combination table with a one-to-one mapping is constructed for the input gain error data, finally forming the logical learning type multi-point piecewise error calibration learning data, which can be used for subsequent real-time error compensation and calibration.

[0020] Step S1 includes the following steps: Step S11: Obtain the operation status log of the current meter; Step S12: Clean the operation status log of the current meter to obtain the cleaned operation status log of the current meter; Step S13: Extract the multi-frequency instantaneous overload current from the cleaned operation status log of the current meter to obtain the multi-frequency instantaneous overload current data; Step S14: Perform waveform time-domain analysis on the multi-frequency instantaneous overload current data to obtain the multi-frequency instantaneous overload current waveform time-domain data.

[0021] In the embodiments of the present invention, the operation of obtaining the operation status log of the current meter is targeted at a precision current measurement meter. The meter is built-in with a 16-bit ADC analog-to-digital converter, and the sampling frequency is 100 kHz. The operation log is recorded in the local memory in CSV format, and the file name is named after the year, month, day and timestamp. The log record includes five types of fields: timestamp (accuracy 0.1 ms), sampled current value (unit mA), temperature value (unit °C), operation status identification code (including working / calibrating / sleeping / abnormal identification, coding range 000-111), and system voltage status (unit V). By setting an automatic task scheduler, the log data within every 24 hours is read periodically, and after being unified and summarized, it is stored in the local server to form a complete original log of the meter operation status. Log cleaning is performed on the above original log, mainly adopting a three-round processing strategy. First, field integrity verification is performed to check whether there are missing fields or format-exception characters in each row. For the record rows with missing or incorrect formats, they are directly excluded. Then, operation status screening is performed, and only the data segments with the operation status identification being "001", that is, in the metering working state, are retained, and the abnormal measurement data segments with the identifications of "000", "010", "100", and "111" are excluded. Finally, based on the temperature and voltage stability verification rules, the record segments with the ambient temperature between 15 °C and 35 °C and the system voltage between 4.8 V and 5.2 V are screened to prevent the unstable segment data caused by high temperature or low voltage from leading to measurement deviation from entering the subsequent analysis. After these three rounds of cleaning processing, the output is the cleaned log of the current meter operation status, and the total number of logs remains above 85% of the original records. For the cleaned log, multi-frequency instantaneous overload current extraction is performed. First, a current mutation extraction threshold is set, and "overload" is defined as an unexpected event where the current value exceeds 100 A (the rated upper limit value of the meter) among three consecutive sampling points. Scan the log according to this rule. At the starting point of each time the condition is met, 50 sampling points (i.e., 0.5 ms) are extracted forward, and 150 sampling points (i.e., 1.5 ms) are extracted backward to form a single instantaneous overload current record with a total duration of 2 ms. Repeat this extraction operation to obtain all the event windows that meet the overload definition in the 24-hour log, and uniquely number and timestamp each data segment. To avoid duplicate statistics, the overload segments with an interval less than 10 ms between events are merged into the same event. The final output format is multiple independent instantaneous overload current data blocks, with each data block having a length of 200 sampling points, and the fields including timestamp, current value sequence, and corresponding voltage and temperature background. Waveform time-domain analysis is performed using the above-extracted multi-frequency instantaneous overload current data. First, each current data sequence is normalized so that its peak value is standardized to 1 for subsequent feature extraction. Subsequently, first-order difference and second-order difference processing are performed on the waveform to calculate the steepness of the rising edge and the falling edge, and the maximum slope value is extracted with a sliding window length of 10 points as the rising steepness index.Meanwhile, the peak duration is calculated by finding the time interval during which the waveform rises from zero to the peak and then falls back to zero, and the peak time point and amplitude are recorded. To extract the local oscillation characteristics of the waveform, each segment of the waveform is decomposed three times using Haar wavelet transform, and the energy value of the third layer coefficients is extracted as the high-frequency perturbation index. All the features form an 8-dimensional vector, including in sequence the peak value, rising steepness, falling steepness, peak duration, starting point position, ending point position, oscillation energy, and average current value. The same-dimensional feature statistics are performed on all events to form a time-domain dataset of multi-frequency instantaneous overload current waveforms, which is used to support subsequent resistance drift simulation and error modeling.

[0022] Step S2 includes the following steps: Step S21: Calculate the overshoot repetition ratio difference of the overload waveform based on the time-domain data of the multi-frequency instantaneous overload current waveform to obtain the overshoot repetition ratio difference of the overload current; Step S22: Conduct simulation analysis on the shunt resistance value drift according to the overshoot repetition ratio difference of the overload current to obtain the shunt resistance value drift data; Step S23: Quantify the drift intensity of the shunt resistance value drift data by piecewise exponential to obtain the piecewise exponential of the resistance value drift intensity; Step S24: Conduct non-linear quantization of the gain error of the preamplifier according to the piecewise exponential of the resistance value drift intensity to obtain the gain error quantization data of the preamplifier.

[0023] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: Calculate the overshoot repetition ratio difference of the overload waveform based on the time-domain data of the multi-frequency instantaneous overload current waveform to obtain the overshoot repetition ratio difference of the overload current; In the embodiment of the present invention, the operation of calculating the overshoot repetition ratio difference of the overload waveform based on the time-domain data of the multi-frequency instantaneous overload current waveform is completed by statistically counting all the peak overshoot times and their amplitude distributions in the current waveform. The specific implementation method is as follows: First, perform a second-order difference on each 2-ms long waveform sequence to locate the local maximum points in the transient response, that is, the overshoot peaks in the waveform. Define "overshoot" as the subsequent secondary peak within 3 ms starting from the main peak, and its amplitude is not less than 10% of the main peak value. Record the main peak value P0 and the subsequent secondary peaks for each segment of the waveform, and calculate the ratio of each secondary peak to the main peak to form a set , named as the overshoot ratio sequence. Subsequently, count the overshoot ratio sequences of all waveforms within one operating cycle of a current meter (such as 24 hours), and calculate the maximum difference in their repetition ratios, that is, the difference between the maximum and minimum values that appear at the same ratio position (for example, the first overshoot ratio). This difference reflects the consistency of the component response under the same overload stimulus, and is named "overshoot repetition ratio difference". For example, in an experiment, among 876 overload waveforms recorded by a certain meter, the average value of the first overshoot ratio is 0.21, the maximum is 0.28, and the minimum is 0.17, then the overshoot repetition ratio difference of the first item is 0.11.

[0024] Step S22: Conduct a shunt resistance value drift simulation analysis based on the overshoot repetition ratio difference of the overload current to obtain shunt resistance value drift data; In the embodiment of the present invention, a shunt resistance value drift simulation analysis is carried out according to the above overshoot repetition ratio difference. The specific method is as follows: First, model the shunt resistance as a temperature rise response element, and based on the typical resistance thermal drift model in the IEC 60115 standard, set the initial nominal resistance value to 100 μΩ. Using the heat accumulation function, combined with the instantaneous peak current and overshoot repetition ratio difference data, calculate the equivalent thermal stress value caused by a single overload event. This thermal stress value is accumulated after each event occurs to form an accumulated thermal stress curve. Then, according to the strain-resistance value drift empirical curve of the material (the measured calibration table provided by the manufacturer), convert the accumulated thermal stress into the resistance value change amount per unit time. For example, in the experimental sample, it is set that the thermal stress unit ( ) corresponds to a resistance value increase of 0.3 μΩ, then the accumulated thermal stress value within one operating cycle is , corresponding to a resistance value drift of 0.96 μΩ. The final output is the resistance value change data in the time series format, and each recording point includes a time stamp, an accumulated thermal stress value, and the current resistance value.

[0025] Step S23: Perform a piecewise exponential quantization on the shunt resistance value drift data to obtain a piecewise exponential of the resistance value drift intensity; In the embodiment of the present invention, a piecewise exponential quantization is performed on the above shunt resistance value drift data. First, divide the resistance value drift data into several segments in units of 1 hour, and calculate the maximum amplitude of the resistance value change within each segment to obtain the drift intensity of each hourly segment. Then, set four-level segmentation thresholds, for example: <0.2 μΩ, 0.2 - 0.5 μΩ, 0.5 - 1 μΩ, >1 μΩ, with corresponding piecewise exponential levels being 1, 2, 3, and 4 respectively. All time periods are classified in segments, and a piecewise index array of resistance value drift intensity is generated. Each segment record includes the start time, end time, resistance value drift amount, and the corresponding drift intensity piecewise exponent. In the experimental data, a certain resistor drifts by 0.34 within the period from 00:00 to 01:00 μΩ, with the assigned index level being 2, and it drifts by 1.1 within the period from 01:00 to 02:00 μΩ, with the assigned index level being 4, and finally a 24 - segment structured intensity piecewise index array is formed.

[0026] Step S24: Perform non - linear quantization of the gain error of the pre - amplifier according to the piecewise exponent of the resistance value drift intensity to obtain the quantization data of the pre - amplifier gain error.

[0027] In the embodiment of the present invention, non - linear quantization of the gain error of the pre - amplifier is performed according to the above - mentioned piecewise exponent of the resistance value drift intensity. The method is to establish a non - linear mapping table between the resistance value drift and the gain error. First, a reference lookup table is established through measured calibration data. The table consists of the resistance change value (unit: μΩ) and the output error of the pre - amplifier (unit: mV), and the interpolation method is cubic spline interpolation. Then, according to the drift intensity index of each segment, the corresponding average resistance change value is extracted (for example, the average drift corresponding to intensity level 2 is 0.35 μΩ), which is input into the above non - linear interpolation function to obtain the gain error value of the current segment. The error values of all segments are accumulated to form a complete error time series, and a third - order moving average smoothing is performed on the error series to eliminate non - physical mutations. The final output is a gain error quantization data table, which records the gain deviation value within each hourly segment and its corresponding drift level. The data structure is a triple sequence of timestamp, gain error value, and drift intensity level for subsequent multi - point calibration and error calibration.

[0028] In another embodiment, in the embodiment of performing non - linear quantization of the gain error of the pre - amplifier according to the piecewise exponent of the resistance value drift intensity, error quantization is achieved by introducing a model of the non - linear influence of the input resistance change on the gain output of the pre - amplifier. The operation process first extracts the resistance value drift range corresponding to each intensity level from the piecewise exponent of the resistance value drift intensity. For example, it is set that intensity segment 1 corresponds to 0.2 μΩ, intensity segment 2 corresponds to 0.2 - 0.5 μΩ, intensity segment 3 corresponds to 0.5 - 1 μΩ, and intensity segment 4 is greater than 1 μΩ. Take the structure with a typical preamplifier input resistance of 10 mΩ and a feedback resistance of 100 mΩ as the analysis object. Using the basic gain expression of the voltage amplifier and combining with the slight change in the input voltage amplitude caused by the change in the front-end shunt resistance, calculate the actual gain change rate. Conduct static simulation calculations on the input voltage drop errors introduced by different resistance value drift values. Set the input current to be constant at 5 A, and the corresponding ideal voltage drop is 500 μV. When the shunt resistance drifts by 0.5 μΩ, the actual voltage drop is 502.5 μV, and the relative deviation from the input voltage is 0.5%. This deviation forms a non-linear gain change within the standard amplifier operating range, and it is necessary to consider its impact on the amplitude expansion of the output, thereby affecting the linear characteristics of the entire measurement chain. To handle this non-linear relationship, use the one-dimensional double-logarithmic coordinate fitting method. Take the resistance value drift as the independent variable and the actual measurement deviation value at the output end as the dependent variable. The sampling points range from 0 μΩ to 2 μΩ, and a sampling point is set every 0.1 μΩ, forming a total of 21 groups of data points. Perform Log-Log fitting on these data to obtain a fitting curve and construct a non-linear gain offset lookup table through the difference algorithm. Convert the resistance value drift intensity index of each segment obtained in the previous step into the interval median value. For example, for segment 3, the drift value is taken as 0.75 μΩ. Through the lookup table, the corresponding gain error is 0.82%. After performing such quantization processing on all intensity segments, generate a set of structured gain error data. Each record item includes the segment number, drift median value, voltage offset amount, and the corresponding gain error percentage value. For example, in a certain current sensor sample, it is recorded that the resistance value drift intensity index at the 10th hour is 3, the mapped median value is 0.75 μΩ, and the corresponding gain error is 0.82%.

[0029] Step S23 includes the following steps: Step S231: Analyze the relative change amount of the shunt resistance value drift data to obtain the relative change amount of the shunt resistance value drift; Step S232: Identify the multi-stage logarithmic gradient of the increasing resistance value of the shunt resistance value drift data according to the relative change amount of the shunt resistance value drift to obtain the multi-stage logarithmic gradient of the increasing resistance value; Step S233: Calculate the growth difference of the discrete monotonic increasing rate for the multi-stage logarithmic gradient of the increasing resistance value to obtain the growth difference of the discrete monotonic increasing rate; Step S234: Fit the extreme value interval of the drift segment intensity for the multi-stage logarithmic gradient of the increasing resistance value based on the growth difference of the discrete monotonic increasing rate to obtain the fitting data of the shunt resistance value drift segment interval intensity; Step S235: Quantize the drift intensity segment index according to the fitting data of the shunt resistance value drift segment interval intensity to obtain the shunt resistance value drift intensity segment index.

[0030] In the embodiments of the present invention, first, the relative change amount of the resistance drift of the shunt resistance data is analyzed. The specific operation method is as follows: Select the resistance value at the reference time point from the measured resistance value sequence of the shunt resistance recorded at different times, and calculate the relative change amount of the measured resistance value at all subsequent time points in the form of sequence difference. The entire sequence is sampled once every 5 minutes, the sample duration is 48 hours, and there are a total of 576 data points. The constant load current is maintained at 10 A during the data acquisition interval, and the ambient temperature is stable at 25 ± 1 °C. After the acquisition is completed, a relative change amount sequence is formed for subsequent analysis. According to the above sequence, multi-stage logarithmic gradient identification of resistance increase is performed. The operation method is as follows: Perform logarithmic transformation on the sequence to generate the sequence, and eliminate Data points ≤0 are used to avoid undefined mathematics. In the log domain, a logarithmic change spectrum is constructed with time as the abscissa. The slope within each window is extracted by the method of sliding linear regression with a fixed time window. The window length is 10 sampling points, and the sliding step is 1 sampling point. A total of 566 groups of local slopes are extracted as the stage logarithmic gradient features, and the slope value, the start and end points of the corresponding time period, and the fitting residual are recorded. These slopes are segmented from small to large according to the absolute value, and each segment is defined as a resistance value change gradient stage. A total of 5 segments are divided, and so on until the fifth segment is [0.001, maximum value], and a multi-stage resistance increase logarithmic gradient mapping relationship table is established. A sequence is constructed based on the logarithmic slopes extracted from each of the above segments, and the discrete monotonic increasing rate growth difference of this sequence is calculated. The specific method is as follows: Define the difference between the slopes of every two adjacent gradient segments as the first-order difference, and then perform the second-order difference to obtain the rate growth change amount. After smoothing the slope values within all segments using spline interpolation and then performing the difference, it is to avoid abnormal recognition caused by sudden rate changes due to noise. Through this process, a data point sequence with a length of 4 can be obtained, reflecting the rate growth trend between each gradient segment. For the data segments showing a continuous positive growth trend, they are marked as monotonic increasing features, and the growth difference of each segment and its corresponding time index range are recorded. The discrete monotonic increasing rate growth difference is used to fit the extreme value interval of the drift segmentation intensity of the stage resistance increase logarithmic gradient. The specific fitting method uses piecewise least squares linear fitting. Each segment corresponds to a slope growth trend interval, and the data points of each segment are approximated in a piecewise approximation manner. The optimal segmentation point of the interval is determined by calculating the sum of the squares of the fitting residuals. The optimal segmentation point is determined based on the minimum residual to form the intensity extreme value interval division. For example, the first interval is 0 - 12 hours, the second interval is 12 - 26 hours, and the third interval is 26 - 48 hours. The maximum slope growth difference within each segment is recorded as the intensity extreme value of this segment. The final output data structure includes fields such as interval number, start and end time points, fitting line slope, maximum rate increase value, etc., which are used for the input of the next-stage exponential quantization. According to the drift segmentation interval intensity extreme value data obtained by fitting, drift intensity segmentation exponential quantization is performed. The specific method is as follows: After normalizing the extreme value data to the 0 - 1 interval range, it is segmented and encoded according to the set 5-level segmentation rule. The segmentation thresholds are set as follows: The first level is 0 - 0.2, the second level is 0.2 - 0.4, and so on. The fifth level is 0.8 - 1.0. The maximum intensity value of each normalized segment is classified according to the interval and given the corresponding exponential label. For example, the intensity value of interval 2 is 0.35, corresponding to the second level, and is marked as the exponent "2". Finally, a structured output data table is formed, including the start and end times corresponding to each drift segment, the normalized intensity value, and the quantized intensity index, which are used as the input data for the subsequent preamplifier error analysis.

[0031] Step S24 includes the following steps: Step S241: Perform multi-scale difference processing on the piecewise exponential of the resistance value drift intensity to obtain the multi-scale difference data of the resistance value drift intensity; Step S242: Based on the multi-scale difference data of the resistance value drift intensity, perform simulation on the distortion of the output voltage signal of the preamplifier to obtain the distortion data of the output voltage signal; Step S243: Analyze the disorder of the non-linear amplification of the output voltage signal distortion data to obtain the non-linear amplification disorder data of the voltage signal; Step S244: Perform time-series dependent signal interpolation processing according to the non-linear amplification disorder data of the voltage signal to obtain the disorder interpolation data of the voltage signal; Step S245: Perform non-linear quantization of the gain error of the preamplifier according to the disorder interpolation data of the voltage signal to obtain the quantization data of the gain error of the preamplifier.

[0032] In the embodiment of the present invention, multi-scale difference processing is performed on the piecewise exponential of the resistance value drift intensity. This process adopts a three-layer nested scale structure, and the basic scale window is set to 5 points, the medium scale window is set to 15 points, and the large scale window is set to 45 points respectively. The processing object is the piecewise exponential sequence of the resistance value drift intensity obtained in step S225, with a total of 72 segments of data. For each scale window, taking the center point as the reference, extract the exponential value sequences in the equal-length regions before and after, and perform first-order difference processing. The difference method is the direct difference between adjacent two points to form the corresponding difference sequence, and then construct the multi-scale difference matrix in turn. This matrix is a three-dimensional structure, and each dimension corresponds to the time period index, the scale category, and the difference result value respectively, and is recorded in a structured array for subsequent processing. The difference result shows the fluctuation rate and direction change trend of the drift index at different time scales, effectively revealing the hidden disturbance of the measurement level fluctuation caused by the resistance value change. Based on the above-obtained multi-scale difference data of the resistance value drift intensity, construct a simulation of the distortion of the output voltage signal of the preamplifier. The simulation adopts a point-by-point mapping method, constructs an exponential gain perturbation function according to the difference data, and the perturbation function is expressed as a linear superposition form of the difference results of each scale. By setting the linear combination weights, the basic scale is set to 0.6, the medium scale is set to 0.3, and the large scale is set to 0.1 to simulate the gain adjustment behavior of the preamplifier under different time response characteristics. The original ideal output voltage is set as a reference linear function , where k is a fixed gain constant (the default value is k = 18), I(t)is the current input signal at the corresponding moment, which is a sinusoidally modulated mixed DC form with a frequency of 1Hz and an amplitude range of 0~10A. The differential disturbance function is mapped to the original voltage output, and the analog gain is amplified non-stable to form an analog voltage output signal sequence. Finally, the output voltage signal distortion data is formed, including the voltage value at each moment, the disturbance superposition factor, and the total distortion offset value, and the data accuracy is maintained at the order of 0.001V. For the analog output voltage signal distortion data, nonlinear amplification disorder analysis is performed. In the operation, the signal normalization processing is first performed to normalize the entire sequence to the [0,1] interval, and the maximum and minimum value normalization method is used to process the original voltage distortion data. Then, the delayed embedding reconstruction method is applied to reconstruct the phase space. The embedding dimension is set to 6 and the delay time is 3 sampling points. After constructing the phase space trajectory, the trajectory sequence is subjected to permutation entropy analysis, with a sliding window of 50 points and a step size of 1 point. The voltage sequence in each window is fully sorted and statistically analyzed, and the corresponding permutation entropy value is calculated. The permutation entropy value range is recorded between 0.42 and 0.89, indicating the disorder degree of the signal after nonlinear amplification. The root mean square fluctuation analysis is further performed on the entropy value sequence to extract the long-range correlation and the local disorder diffusion trend, so as to judge the nonlinear gain response characteristics corresponding to the output signal in the specific resistance drift stage. The nonlinear amplification disordered data of the voltage signal finally output includes: the permutation entropy value of each time window, the embedding dimension parameter, the corresponding analog voltage value, the distortion offset range and its propagation effect under the gain perturbation function. The time-dependent signal interpolation processing is performed according to the nonlinear amplification disordered data of the voltage signal. The processing process is based on the nonlinear amplification disordered data of the voltage signal obtained in step S243, and the recorded content includes the normalized voltage signal value, the corresponding permutation entropy value and the nonlinear perturbation amplitude. The signal has a short-term fluctuation interval caused by gain disturbance and a nonlinear jump point within the sampling time. It needs to be finely interpolated and reconstructed to restore the time series continuity characteristics and eliminate the disordered noise error. The interpolation method uses a combination of segmented cubic spline interpolation and Savitzky-Golay smoothing filtering. The spline interpolation uses every 0.01s as a segment node, extracts 10 points of each segment to construct a local cubic polynomial interpolation function, and the interpolation result covers all data points in the original time domain and interpolates and compensates for the uneven sampling segment; at the same time, the Savitzky-Golay filter is performed on the spline output area with the local maximum second-order derivative abnormal mutation. The window width is set to 11 points and the polynomial order is set to 3rd order. After filtering, the curvature fluctuation of the signal is weakened and the continuity of the time series is improved. The interpolation output data includes the normalized voltage value after interpolation, the interpolation residual, the local fluctuation intensity and the corresponding time index, which constitute the complete disordered interpolation data of the voltage signal. The nonlinear quantization of the gain error of the preamplifier is performed based on the disordered interpolation data of the voltage signal obtained above. The goal of this step is to extract the gain nonlinear response characteristics of the preamplifier caused by resistance drift under actual usage conditions.The specific quantification method is to establish a theoretical linear relationship reference curve between the normalized input current and the interpolated output voltage, and calculate the deviation error between the interpolated output and the reference. The reference curve is constructed based on an ideal linear amplification model, where the input current range is from 0 A to 10 A, corresponding to a normalized input of 0 to 1. A fixed gain constant k = 18 is set, and the reference output is a linear function. Vout = k × I, Vout: represents the output voltage of the amplifier (Voltage Output), in volts (V). It is the output result after the current signal passes through the preamplifier. This value is the final voltage signal for the instrument to measure or the recording system to read. I: represents the input current signal, in amperes (A). It is the original current signal flowing through the shunt resistor or the object under test. The interpolated signal comes from the output of step S244. Compare the theoretical voltage and the interpolated voltage point by point, and calculate the relative error rate. . Among them, : represents the voltage signal value after interpolation processing, that is, the voltage output data restored or corrected by the time-series dependent signal interpolation algorithm; : represents the theoretically linear voltage signal value, usually the voltage output calculated based on the ideal linear amplifier model, that is Vout , representing the ideal reference voltage without gain error or nonlinear distortion; Perform nonlinear feature analysis on the entire signal error sequence. Use the changing trends of the first derivative and the second derivative of the error to construct an error gradient sequence. By extracting the distribution densities of the local maximum positive error interval and the local maximum negative error interval, divide the error quantification interval. Each interval corresponds to a different gain nonlinear response area. Finally, form a gain error quantification data structure, including: the distribution map of the absolute value range of the error, the index of the nonlinear response interval, the integral density distribution of the error, the index of the maximum deviation point, and the corresponding timestamp.

[0033] Step S244 includes the following steps: Calculate the average difference in the signal change amplitudes before and after the time series for the disordered data of the nonlinearly amplified voltage signal to obtain the average difference in the signal change amplitudes of the time series; Perform signal change variational inference processing based on the average difference in the signal change amplitudes of the time series to obtain approximate data on the signal change distribution difference; Perform marginal likelihood estimation on the approximate data of the signal change distribution difference to obtain the marginal likelihood estimation data of the change difference; Perform time-series dependent signal interpolation processing on the marginal likelihood estimation data of the change difference based on Bayesian interpolation to obtain disordered interpolation data of the voltage signal.

[0034] In the embodiment of the present invention, first, obtain a sequence of disordered data of the nonlinearly amplified voltage signal arranged according to the sampling time series. Assume that the original data sampling frequency is 10 kHz and the recording length is 100 ms, that is, a total of 1000 sampling points are included, denoted as the sequenceV_n (n ∈ [1, 1000]) For any two adjacent sampling points in the sequence V_n and V_{n + 1} , calculate the absolute value of their difference, denoted as ΔV_n = |V_{n + 1} - V_n| , which is used to measure the amplitude change of the voltage before and after this time point. Summing up all Δ V_n and then dividing by the total number of points, the average difference in the amplitude change of the time series signal for the entire sampling period is obtained, which is used to characterize the irregular fluctuation intensity of the voltage signal under non-linear amplification distortion conditions, forming the basic data for the statistical analysis of the local distortion time series change behavior. Based on the above average difference data, further variational inference processing of the signal change is performed. By performing kernel density estimation on the probability distribution density of the ΔV_n data sequence, an approximate change probability distribution function is obtained, and on the basis of this distribution, a variational distribution function Q(x) is constructed. By minimizing the Kullback-Leibler divergence, an approximation to the true distribution P(x) is achieved. This processing adopts a sliding window mechanism with a fixed window width of 10 points. In each window, the distribution offset is estimated based on local statistical features, thereby obtaining approximate data on the distribution difference of the signal change, which is used to describe the deviation degree between the non-linear fluctuation region and the normal region at the statistical level. Next, a marginal likelihood estimation operation is performed on the above approximate data on the change distribution difference. By calculating the integral of the likelihood function of the observed value under the variational distribution in each sliding window, the marginal likelihood value at this point is obtained, forming a continuous marginal likelihood estimation data sequence. This sequence still maintains consistency with the original sampling data in the time dimension, providing a basis for time consistency during subsequent interpolation reconstruction. To improve the estimation accuracy of the marginal likelihood, a Bessel correction factor is introduced to correct the variance estimation to reduce the bias under small sample windows. Finally, based on the above marginal likelihood estimation data, Bayesian interpolation is used to perform time-series dependent interpolation processing on missing, scrambled, or disordered voltage signal data segments. During the interpolation processing, a Gaussian process prior model is selected, the kernel function uses a squared exponential kernel, and the hyperparameters are obtained by maximizing the marginal likelihood. The interpolation value V_interp(t) at each moment t is given by the predicted mean of the Gaussian process. After interpolating and reconstructing multiple segments of abnormal points, voltage signal disordered interpolation data with the same length as the original data and the same sampling interval is obtained for subsequent gain error quantization analysis. The input data range for the interpolation processing is low-frequency signals between 0 - 5V, the typical error bandwidth does not exceed 0.01V, and the interpolation density is maintained at 1 point inserted every 10 μs to ensure the time consistency and continuity of the final data structure.

[0035] Step S3 includes the following steps: Step S31: Analyze the trend of the time series error increment of the preamplifier gain error quantization data to obtain the time series error increment trend data; Step S32: Perform multi-point calibration of error variation based on the trend data of the timing error increment to obtain the multi-point calibration data of the error increment variation; Step S33: Perform multi-point segmented error calibration based on the multi-point calibration data of the error increment variation to obtain the multi-point segmented error calibration data; Step S34: Perform logical learning on the multi-point segmented error calibration data to obtain the multi-point segmented error calibration learning data.

[0036] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: Analyze the trend of the timing error increment for the preamplifier gain error quantization data to obtain the trend data of the timing error increment; In the embodiment of the present invention, the preamplifier gain error quantization data is sorted according to the time stamp to construct an equally spaced time series. The sampling period is set to 20 μs, the total duration is 200 ms, and the total number of data points collected is 10,000. Each point of error quantization data is in millivolts and is marked as E_t (t is the time index). By calculating the difference between the error values of two consecutive time points ΔE_t = E_{t + 1} - E_t , the error increment sequence is obtained. To eliminate the influence of short-term fluctuations, a 5-point moving average method is used to smooth the error increment sequence. Subsequently, a first derivative analysis method is used to extract the overall trend characteristics of the error over time. For the trend changes of error increase or decrease in a long period, local linear regression (LOWESS) is further used for trend fitting to extract the complete trend data of the timing error increment, which is used to describe the change path of the preamplifier error accumulated over time, with the unit of millivolt per second.

[0037] Step S32: Perform multi-point calibration of error variation based on the trend data of the timing error increment to obtain the multi-point calibration data of the error increment variation; In the embodiment of the present invention, based on the error increment trend data obtained in S31, several typical error change points are selected as multi-point calibration reference points. The specific selection method uses the extreme value sampling method to extract the local maximum and minimum points of the error increment trend curve, and an equally spaced gain error segment is set as the boundary interval, such as every 50 mV as a segment, and the error interval is divided into 0–50 mV, 50–100 mV, 100–150 mV, etc. A representative point is selected as the calibration reference value within each segment, and the number of calibration points is not less than 10 in actual tests. By introducing known standard current input values, such as 5 mA, 10 mA, 20 mA, 30 mA, at each calibration point, measuring the amplifier output and comparing it with the theoretical linear output, and recording the actual error value as the error increment value corresponding to the calibration point. Finally, the multi-point calibration data of the error increment variation is obtained, which constitutes the segmented basis for the subsequent calibration operation.

[0038] Step S33: Perform multi-point segmented error calibration based on the multi-point calibration data of error increment change to obtain multi-point segmented error calibration data; In the embodiment of the present invention, the gain error is segmented and calibrated by using the above multi-point calibration data. The entire gain error change interval is divided into several segments, and each segment corresponds to a calibrated error compensation function. In specific operations, a linear interpolation method is used to construct the error compensation function between two adjacent calibration points. If the error interval span is large, spline interpolation is used instead to ensure the compensation accuracy. An error mapping relationship is independently established for each section, with the input being the error difference between the measured value and the theoretical value, and the output being the required compensation correction amount. By inputting the measured voltage signal error into the error compensation function of the corresponding section, the error correction value is output and added to the original value to achieve the calibrated result after error correction. The entire calibration process is executed one by one for all time points, and the completed result is the multi-point segmented error calibration data, whose data format is consistent with the original error quantization data, facilitating subsequent logical analysis.

[0039] Step S34: Perform logical learning on the multi-point segmented error calibration data to obtain multi-point segmented error calibration learning data.

[0040] In the embodiment of the present invention, for the calibrated data, its logical structure and variation law are further extracted. The decision tree learning method is used to classify and summarize the multi-point segmented error calibration data. The input features corresponding to each error calibration data point are defined as: timestamp t, original error value E_t , compensated value E'_t and their difference ΔE_t = E'_t - E_t . An input-output training set is constructed for all samples, and a tree structure is constructed by maximizing the information gain. A threshold judgment condition is set at each node, and the output is the logical rule for the corresponding error correction. To improve the generalization ability, 10-fold cross-validation is used to adjust the model structure, and the feature importance is sorted. After removing redundant features, the logical path is optimized. The output result is the multi-point segmented error calibration learning data, which includes the error variation law description rules and corresponding logical expressions for each section, and is stored in a rule table structure, which can be used for further automatic calibration strategy optimization.

[0041] Step S33 includes the following steps: Step S331: Perform error mutation difference operation on the multi-point calibration data of error increment change to obtain multi-point error mutation difference data; Step S332: Perform multiple regression analysis on the multi-point error mutation difference data to obtain multi-point error mutation regression data; Step S333: Perform non-constant variance calculation of multi-point error mutation based on the multi-point error mutation regression data to obtain non-constant variance of multi-point error mutation; Step S334: performing time series multi-point error structure similarity analysis on the error multi-point mutation regression data to obtain time series multi-point error structure similarity data; Step S335: performing error equal compensation value change setting according to the error multi-point mutation non-constant variance and the time series multi-point error structure similarity data to obtain error equal compensation value setting data; Step S336: Perform multi-point segmented error calibration based on the error equal compensation value setting data to obtain multi-point segmented error calibration data.

[0042] In the embodiment of the present invention, the obtained error increment change multi-point calibration data are first arranged in order from small to large according to the current input value. For each point in the sequence, the error difference between it and the previous calibration point is calculated to form a differential sequence. The current input value interval of each calibration point is set to 5mA, and a total of 12 calibration points are set, with a current coverage range of 5mA to 60mA. The error calibration value corresponding to each point is in millivolts, and the form of differential operation is . A complete error mutation differential data sequence is obtained through continuous calculation, and the statistical discrimination method is used to identify the points where the mutation amplitude is greater than the mean plus two times the standard deviation, which are recorded as the error mutation critical point. The above error multi-point mutation differential data is used as the dependent variable, and the current input value I, the voltage output reference value V_ref, and the sampling time t are used as independent variables to construct a multivariate regression analysis model. The least squares method is used to solve the coefficients and fit the relationship between the error mutation differential value and the three-dimensional input features. To improve robustness, a white noise hypothesis test is performed on the regression residual, outliers are removed, and the model coefficients are refitted. Finally, the error multi-point mutation regression data is generated, and the data structure is ,in ΔE_pred It is the predicted value of the fitting error mutation under the current input conditions, in millivolts. The regression data is used to characterize the parameterized trend of the mutation error, which is convenient for subsequent variance and structure analysis. Based on the error multi-point mutation regression data, the heteroscedasticity test method is used to perform regression analysis on the residual square value and the independent variable of each fitting point to determine whether the error fluctuates non-constantly with the change of the input variable. In the specific operation, the Breusch-Pagan test is used to evaluate the regression R² of the regression residual square and the independent variable, and the significance level is set to 0.05. When the test statistic is significant, it is considered that there is a non-constant variance phenomenon. For each error point, the variance change with current and time is calculated and recorded as the error multi-point mutation non-constant variance data. The data structure is {I, t, σ²} ,in σ²is the local error variance estimation value. This data is used to quantify the error uncertainty under different input conditions and helps to construct the section correction weight factor. Perform a structural similarity comparison on each set of time-series data segments in the error multi-point mutation regression data. Specifically, construct an error vector with every 10 consecutive time points as a group, and calculate the cosine similarity and Euclidean distance between pairwise error vectors. Using the similarity greater than 0.95 and the distance less than 0.05 as the judgment criteria, classify them into one type of error structure segment. After dimensionality reduction of each structure segment through principal component analysis (PCA), perform a clustering operation, and finally obtain several groups of time-series error structure similarity clusters, with each cluster representing a type of similar error behavior. The output data is time-series multi-point error structure similarity data, and its content includes key features such as clustering numbers, the time period of the affiliated points, mean error, and error variance. Combining the non-constant variance data obtained in step S333 and the structural similarity results in step S334, set the error compensation values within each structure cluster. First, calculate the average error value μ and the average variance σ² within each cluster. Then, set an error amplification factor γ>1 for the clusters with larger variances and γ<1 for those with smaller variances, and ensure that the compensation operation is proportional to the error fluctuation intensity by adjusting the compensation amplitude. The setting of the compensation value is based on the structure cluster as a unit, and calculate the equal compensation constant k_i = μ_i × γ_i for each cluster as the unified compensation value for all points within the cluster. Finally, generate the error equal compensation value setting data for guiding error correction. Map the error equal compensation value setting data to the original multi-point error sequence. For each error point, extract the corresponding compensation value k_i according to the structure similarity cluster it belongs to, and perform weighted correction on the original error value, that is E'_i = E_i + k_i . This process is strictly carried out according to the structure cluster division and compensation value mapping relationship. After all compensations are completed, the output corrected error data is the final multi-point segmented error calibration data.

[0043] Preferably, the present invention also provides an instrument measurement and detection calibration system for performing the instrument measurement and detection calibration method described above. This instrument measurement and detection calibration system includes: A waveform time-domain analysis module for obtaining the operation status log of the current meter; extracting multi-frequency instantaneous overload currents from the operation status log of the current meter, and then performing waveform time-domain analysis to obtain multi-frequency instantaneous overload current waveform time-domain data; An error non-linear quantization module for performing shunt resistance value drift simulation analysis based on the multi-frequency instantaneous overload current waveform time-domain data to obtain shunt resistance value drift data; performing drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain the resistance value drift intensity segmented index; performing non-linear quantization of the gain error of the preamplifier according to the resistance value drift intensity segmented index to obtain preamplifier gain error quantization data; The multi-point segmented error calibration module is used to perform multi-point calibration of error variation based on the preamplifier gain error quantization data to obtain multi-point calibration data of error increment variation; and perform multi-point segmented error calibration based on the multi-point calibration data of error increment variation to obtain multi-point segmented error calibration learning data.

[0044] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for instrument measurement detection and calibration, characterized in that, It includes the following steps: Step S1: Obtain the operation status log of the current meter; Extract the multi-frequency instantaneous overload current from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain the multi-frequency instantaneous overload current waveform time-domain data; Step S2: Based on the multi-frequency instantaneous overload current waveform time-domain data, perform simulation analysis on the shunt resistance value drift to obtain the shunt resistance value drift data; perform drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain the resistance value drift intensity segmented index; According to the resistance value drift intensity segmented index, perform non-linear quantization of the gain error of the preamplifier to obtain the preamplifier gain error quantization data; Step S3: According to the preamplifier gain error quantization data, perform multi-point calibration of the error change to obtain the multi-point calibration data of the error increment change; Based on the multi-point calibration data of the error increment change, perform multi-point segmented error calibration to obtain the multi-point segmented error calibration learning data.

2. The instrument metering detection and calibration method according to claim 1, characterized in that Step S1 includes the following steps: Step S11: Obtain the operation status log of the current meter; Step S12: Clean the operation status log of the current meter to obtain the cleaned operation status log of the current meter; Step S13: Extract the multi-frequency instantaneous overload current from the cleaned operation status log of the current meter to obtain the multi-frequency instantaneous overload current data; Step S14: Perform waveform time-domain analysis on the multi-frequency instantaneous overload current data to obtain the multi-frequency instantaneous overload current waveform time-domain data.

3. The instrument measurement and detection calibration method according to claim 1, wherein, Step S2 includes the following steps: Step S21: Calculate the overshoot repetition ratio difference of the overload waveform based on the multi-frequency instantaneous overload current waveform time-domain data to obtain the overshoot repetition ratio difference of the overload current; Step S22: According to the overshoot repetition ratio difference of the overload current, perform simulation analysis on the shunt resistance value drift to obtain the shunt resistance value drift data; Step S23: Perform drift intensity segmented exponential quantization on the shunt resistance value drift data to obtain the resistance value drift intensity segmented index; Step S24: According to the resistance value drift intensity segmented index, perform non-linear quantization of the gain error of the preamplifier to obtain the preamplifier gain error quantization data.

4. The instrument metrology detection and calibration method according to claim 3, characterized in that, Step S23 includes the following steps: Step S231: Analyze the relative change amount of the shunt resistance value drift data to obtain the relative change amount of the shunt resistance value drift; Step S232: Based on the relative change amount of the shunt resistance value drift, identify the multi-stage logarithmic gradient of the shunt resistance value increase for the shunt resistance value drift data to obtain the multi-stage logarithmic gradient of the shunt resistance value increase; Step S233: Calculate the growth difference of the discrete monotonic increasing rate for the multi-stage logarithmic gradient of the shunt resistance value increase to obtain the growth difference of the discrete monotonic increasing rate; Step S234: Based on the growth difference of the discrete monotonic increasing rate, perform drift segmented interval intensity fitting on the stage logarithmic gradient of the shunt resistance value increase to obtain the fitting data of the shunt resistance value drift segmented interval intensity; Step S235: According to the fitting data of the shunt resistance value drift segmented interval intensity, perform drift intensity segmented exponential quantization to obtain the resistance value drift intensity segmented index.

5. The instrument measurement and detection calibration method according to claim 3, characterized in that, Step S24 includes the following steps: Step S241: Perform multi-scale difference processing on the resistance value drift intensity segmented index to obtain the multi-scale difference data of the resistance value drift intensity; Step S242: Based on the multi-scale differential data of the resistance drift intensity, perform distortion simulation on the output voltage signal of the preamplifier to obtain the distortion data of the output voltage signal; Step S243: Analyze the disorder of the non-linear amplification of the output voltage signal distortion data to obtain the non-linear amplification disorder data of the voltage signal; Step S244: Perform time-series dependent signal interpolation processing on the non-linear amplification disorder data of the voltage signal to obtain the disorder interpolation data of the voltage signal; Step S245: Perform non-linear quantization of the gain error of the preamplifier based on the disorder interpolation data of the voltage signal to obtain the gain error quantization data of the preamplifier.

6. The instrument metrology detection and calibration method according to claim 4, characterized in that Step S244 includes the following steps: Calculate the average difference of the signal change amplitudes before and after the time series for the non-linear amplification disorder data of the voltage signal to obtain the average difference of the signal change amplitudes in the time series; Perform variational inference processing on the signal change based on the average difference of the signal change amplitudes in the time series to obtain approximate data of the signal change distribution difference; Perform marginal likelihood estimation on the approximate data of the signal change distribution difference to obtain the marginal likelihood estimation data of the change difference; Perform time-series dependent signal interpolation processing on the marginal likelihood estimation data of the change difference based on the Bayesian interpolation method to obtain the disorder interpolation data of the voltage signal.

7. The instrument measurement and detection calibration method according to claim 1, wherein Step S3 includes the following steps: Step S31: Analyze the trend of the time-series error increment of the gain error quantization data of the preamplifier to obtain the time-series error increment trend data; Step S32: Perform multi-point calibration of the error change based on the time-series error increment trend data to obtain the multi-point calibration data of the error increment change; Step S33: Perform multi-point segmented error calibration based on the multi-point calibration data of the error increment change to obtain the multi-point segmented error calibration data; Step S34: Perform logical learning on the multi-point segmented error calibration data to obtain the multi-point segmented error calibration learning data.

8. The instrument measurement and detection calibration method according to claim 7, wherein, Step S33 includes the following steps: Step S331: Perform error mutation difference operation on the multi-point calibration data of the error increment change to obtain the multi-point mutation difference data of the error; Step S332: Perform multiple regression analysis on the multi-point mutation difference data of the error to obtain the multi-point mutation regression data of the error; Step S333: Calculate the non-constant variance of the multi-point mutation of the error based on the multi-point mutation regression data of the error to obtain the non-constant variance of the multi-point mutation of the error; Step S334: Perform time-series multi-point error structure similarity analysis on the multi-point mutation regression data of the error to obtain the time-series multi-point error structure similarity data; Step S335: Set the numerical change of the error equal compensation according to the non-constant variance of the multi-point mutation of the error and the time-series multi-point error structure similarity data to obtain the numerical setting data of the error equal compensation; Step S336: Perform multi-point segmented error calibration based on the numerical setting data of the error equal compensation to obtain the multi-point segmented error calibration data.

9. An instrument measurement detection and calibration system, characterized in that, For implementing the instrument metrology detection and calibration method as described in claim 1, the instrument metrology detection and calibration system includes: A waveform time-domain analysis module, configured to obtain the operation status log of the current meter; extract the multi-frequency instantaneous overload current from the operation status log of the current meter, and then perform waveform time-domain analysis to obtain the multi-frequency instantaneous overload current waveform time-domain data; Error non-linear quantization module, which is used to perform shunt resistor value drift simulation analysis based on the time-domain data of the multi-frequency instantaneous overload current waveform to obtain shunt resistor value drift data; perform drift intensity segmented exponential quantization on the shunt resistor value drift data to obtain the segmented exponential of the resistance value drift intensity; perform gain error non-linear quantization of the preamplifier according to the segmented exponential of the resistance value drift intensity to obtain preamplifier gain error quantization data; Multi-point segmented error calibration module, which is used to perform multi-point calibration of error changes based on the preamplifier gain error quantization data to obtain multi-point calibration data of error increment changes; perform multi-point segmented error calibration based on the multi-point calibration data of error increment changes to obtain multi-point segmented error calibration learning data.

Images