An electric energy metering precision calibration system and method

Abstract

The application relates to the technical field of electric power metering, and discloses an electric energy metering precision calibration system and method. The method carries out pretreatment on collected original electric signals to generate standard electric signal data. A calibration state identifier is generated through dynamic evaluation, and when fine calibration is required, the data is adaptively segmented into electric signal segments and independent verification and filtering are carried out. The electric parameter values of the filtered segments are extracted, and the electric parameter values are divided into normal and abnormal parameter clusters according to the positive and negative deviations from the reference values, and after local sorting, the parameter values are paired into to-be-verified parameter groups. The pre-built parameter correlation model is used to analyze the correlation of each group, the number of effective correlation groups is counted, and the overall calibration index is calculated in combination with the unpaired parameter values, so that whether to start calibration is determined. The method can accurately locate the signal period requiring calibration, reduce invalid calculation, and improve the accuracy and robustness of the calibration decision through parameter correlation analysis.

Description

Technical Field

[0001] This invention relates to the field of power metering technology, specifically to a precise calibration system and method for power metering. Background Technology

[0002] In modern society, electricity metering is a core component of power system operation, playing a vital role akin to the heart in the human body. From daily residential electricity consumption to large-scale industrial production, electricity metering is ubiquitous. Accurate electricity metering is the cornerstone of electricity trade settlement, ensuring fairness and impartiality in electricity transactions. For power companies, accurate electricity metering directly impacts their economic benefits and operating cost accounting, influencing the formulation of power generation plans, the optimization of grid dispatch, and the rational allocation of power resources. If metering errors occur, power companies may face incorrect electricity calculations, leading to revenue losses or increased costs. For users, accurate electricity metering protects their legitimate rights, allowing them to clearly understand their electricity consumption, plan their electricity usage rationally, and avoid overpaying due to metering errors.

[0003] Common traditional methods for calibrating electricity meters include power calibration and pulse calibration. The basic principle of power calibration is to inject a known power signal into the electricity metering device using a standard power source, and then compare the measured power value with the standard power value to determine the accuracy of the metering device and perform corresponding calibration. For example, in practice, the output of a standard power source is connected to the input of the electricity meter, and a specific power value, such as 1000 watts, is set. The power value displayed on the meter is observed; if it displays 1020 watts, there is an error, and the meter needs adjustment and calibration. Pulse calibration uses standard pulse signals to calibrate the electricity metering device. Correction is achieved by measuring the deviation between the standard pulse and the pulse sequence of the meter being calibrated. In practice, a pulse generator sends a series of standard pulse signals, which are received and compared with the pulse signal generated by the meter being calibrated to calculate the deviation, and then the meter is adjusted accordingly.

[0004] Power calibration methods have limitations in terms of accuracy. Due to the limitations of their measurement principles and the inherent precision of the equipment, they struggle to meet the ever-increasing demand for high-precision metering. In applications requiring extremely high accuracy in electricity metering, such as research laboratories and high-end manufacturing, errors in power calibration methods can significantly impact experimental results or production processes. While pulse calibration methods have advantages in some aspects, their applicability is narrow, only applicable to certain types of electricity metering devices. They are ineffective for complex and novel electricity metering systems. Furthermore, pulse calibration methods are slow and inefficient when processing small signals, failing to meet the need for rapid calibration. When calibrating a large number of electricity metering devices, they consume significant time and manpower.

[0005] Aging of internal components in electricity metering devices is a significant internal factor contributing to errors. As usage time increases, the performance of components such as resistors and capacitors within the device gradually changes; for example, resistance values ​​may increase or decrease, and capacitance values ​​may deviate. This affects the transmission and processing of electrical signals, leading to metering errors. Chip performance variations are equally significant. Minor process differences may exist during chip manufacturing, and even chips from the same batch may not exhibit completely consistent performance. These performance variations affect the chip's sampling, calculation, and processing of electrical signals, ultimately resulting in errors in electricity metering.

[0006] Ambient temperature has a significant impact on the accuracy of electricity metering. Temperature changes can alter the parameters of electronic components within the metering device. When the temperature rises, the resistance of some resistors increases, thus changing the current and voltage distribution in the circuit and affecting the accuracy of electricity metering. Humidity is also a critical factor; excessive humidity can cause electronic components to become damp, leading to decreased insulation performance, leakage, and other problems that interfere with the normal transmission and measurement of electrical signals, resulting in metering errors. Furthermore, the complex electromagnetic environment of modern society means that surrounding electromagnetic interference, such as electromagnetic radiation from communication base stations, large motors, transformers, and other equipment, can couple into the circuitry of the electricity metering device, interfering with the electrical signal and causing inaccurate metering. Summary of the Invention

[0007] The purpose of this invention is to provide a precise calibration system and method for electricity metering to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides a method for accurate calibration of electricity metering, the method comprising:

[0009] The raw electrical signals of the power metering device to be calibrated are acquired, and the raw electrical signals are subjected to multi-stage signal preprocessing to generate standard electrical signal data.

[0010] The standard electrical signal data is dynamically evaluated. By comparing the measured signal characteristics with the ideal signal characteristics, a calibration status identifier is generated. When the calibration status identifier requires fine calibration, the standard electrical signal data is adaptively divided into multiple electrical signal segments, and each electrical signal segment is independently verified. Based on the verification results, qualified electrical signal segments are filtered out.

[0011] Extract all electrical parameter values ​​from the remaining electrical signal segments after filtering, and calculate the deviation of each electrical parameter value from the corresponding reference parameter value. Based on the positive and negative characteristics of the deviation value, divide the electrical parameter values ​​into normal parameter clusters and abnormal parameter clusters.

[0012] A local sorting operation is performed on the electrical parameter values ​​in the normal parameter cluster and the abnormal parameter cluster to form an ordered parameter sequence, and adjacent electrical parameter values ​​in the sequence are paired as a parameter group to be verified.

[0013] A pre-built parameter association model is used to perform correlation analysis on each parameter group to be verified, output the association score, and count the number of valid association groups with scores exceeding the threshold.

[0014] By combining the number of effective associated groups and the electrical parameter values ​​of those not involved in pairing, the overall calibration index is calculated, and the calibration process is determined based on the comparison between the overall calibration index and the dynamic calibration threshold.

[0015] Preferably, when acquiring the raw electrical signal of the energy metering device to be calibrated and performing multi-stage signal preprocessing on the raw electrical signal to generate standard electrical signal data, the process includes:

[0016] The original electrical signal is acquired in real time by a high-precision sensor, and the noise of the original electrical signal is eliminated by applying a wavelet transform algorithm to obtain a preliminary purified signal.

[0017] The preliminary purified signal is subjected to amplitude normalization processing to scale the signal amplitude to a standard range, and a digital filter is used to remove high-frequency interference components.

[0018] The processed signal is input into the signal quality evaluation unit to calculate the signal-to-noise ratio and stability index. If the signal-to-noise ratio is higher than the preset threshold and the stability index meets the requirements, the standard electrical signal data is output.

[0019] If the signal-to-noise ratio or stability index does not meet the standard, adjust the acquisition parameters and reacquire the original electrical signal.

[0020] Preferably, when dynamically evaluating the standard electrical signal data and generating a calibration status identifier by comparing the measured signal characteristics with the ideal signal characteristics, the process includes:

[0021] Key signal feature points, including peak value, mean value and phase information, are extracted from the standard electrical signal data and matched point by point with ideal signal feature points in the database.

[0022] A calibration pass flag is generated when the errors between all measured signal feature points and ideal signal feature points are below the tolerance range.

[0023] A calibration failure flag is generated when the errors between all measured signal feature points and ideal signal feature points exceed the tolerance range.

[0024] When the error of some measured signal feature points is within the tolerance range but exceeds it in some cases, a fine calibration mark is generated.

[0025] Preferably, when adaptively segmenting the standard electrical signal data into multiple electrical signal segments, independently verifying each electrical signal segment, and filtering out qualified electrical signal segments based on the verification results, the process includes:

[0026] Based on the characteristics of the signal time series, the standard electrical signal data is divided into overlapping electrical signal segments using a sliding window algorithm, with each electrical signal segment containing a fixed number of sampling points;

[0027] For each electrical signal segment, calculate the consistency score of the signal characteristics within the segment. If the consistency score is higher than the qualified threshold, the electrical signal segment is deemed qualified and filtered out, and only electrical signal segments with a consistency score lower than the qualified threshold are retained.

[0028] Preferably, when extracting all electrical parameter values ​​from the remaining electrical signal segments after filtering, calculating the deviation of each electrical parameter value from the corresponding reference parameter value, and classifying the electrical parameter values ​​into normal parameter clusters and abnormal parameter clusters based on the positive or negative characteristics of the deviation values, the process includes:

[0029] A feature extraction algorithm is used to extract values ​​from each remaining electrical signal segment, including the effective voltage value, current harmonic content, and power factor;

[0030] For each electrical parameter value, the corresponding reference parameter value is obtained by querying the standard parameter table, and the absolute deviation value is calculated. If the absolute deviation value is positive or zero, the electrical parameter value is classified into the normal parameter cluster and the positive deviation amount is recorded. If the absolute deviation value is negative, the electrical parameter value is classified into the abnormal parameter cluster and the negative deviation amount is recorded.

[0031] The normal parameter cluster and the abnormal parameter cluster are subgrouped according to the source of the electrical signal segment, forming multiple parameter subsets.

[0032] Preferably, when performing a local sorting operation on the electrical parameter values ​​in the normal parameter cluster and the abnormal parameter cluster to form an ordered parameter sequence, and pairing adjacent electrical parameter values ​​in the sequence into a parameter group to be verified, the process includes:

[0033] Arrange the electrical parameter values ​​in each parameter subset of the normal parameter cluster in ascending order to generate a normal ordered sequence, and combine two consecutive electrical parameter values ​​in the sequence into a normal parameter group to be verified.

[0034] The electrical parameter values ​​in each subset of the abnormal parameter cluster are sorted in descending order to generate an abnormal ordered sequence, and two consecutive electrical parameter values ​​in the sequence are combined into an abnormal parameter group to be verified.

[0035] Isolated electrical parameter values ​​that do not belong to any parameter subset are not paired and are retained as unassociated parameters.

[0036] Preferably, when performing correlation analysis on each parameter group to be verified using a pre-built parameter association model, outputting an association score, and counting the number of valid association groups with scores exceeding a threshold, the process includes:

[0037] Collect electrical parameter values, reference parameter values, and deviation data from historical calibration tasks, construct a training dataset, and divide it into a training subset and a test subset;

[0038] Initialize the graph neural network as a parameter association model, use a training subset to iteratively train the model, and optimize the model parameters through backpropagation;

[0039] After each iteration, the model's association prediction accuracy is evaluated using a test subset. If the accuracy improvement is less than the minimum increment, the learning rate is adaptively adjusted.

[0040] When the accuracy converges or reaches the maximum number of iterations, the model parameters are fixed to obtain the trained parameter correlation model.

[0041] Input the normal and abnormal parameter groups to be verified into the model, output the association probability value of each group, and if the probability value is greater than the association threshold, it is determined to be a valid association group and the number is accumulated.

[0042] Preferably, when calculating the overall calibration index by combining the number of effective association groups and the electrical parameter values ​​that did not participate in pairing, the following steps are included:

[0043] Calculate the logarithmic transformation value of the number of effective association groups, and multiply it by the total deviation range of the normal parameter cluster and the total deviation range of the abnormal parameter cluster to obtain the association contribution component;

[0044] For electrical parameter values ​​that are not paired, calculate their residual deviations from the corresponding reference parameter values, and sum them to obtain the non-associated contribution components;

[0045] The overall calibration index is derived by summing the associated and unassociated contribution components and multiplying them by a weighting factor determined according to the type of metering device.

[0046] Preferably, when determining whether to initiate the calibration process based on a comparison between the overall calibration index and the dynamic calibration threshold, the process includes:

[0047] Real-time monitoring of environmental factors and workload; dynamic adjustment of calibration thresholds.

[0048] If the overall calibration index is greater than or equal to the dynamic calibration threshold, a calibration command is generated and the automatic calibration mechanism is activated.

[0049] If the overall calibration index is less than the dynamic calibration threshold, a maintenance state instruction is generated, and the calibration process is skipped.

[0050] Preferably, the present invention also includes a precise calibration system based on electricity metering, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-described precise calibration method based on electricity metering.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] In the process of electricity metering calibration, multi-stage signal preprocessing of the raw electrical signal is of great significance. The raw electrical signal is collected from the electricity metering device to be calibrated, and it often contains various complex interference signals, which may originate from the surrounding electromagnetic environment, noise from internal components of the device, etc. Multi-stage signal preprocessing is like finely processing messy raw materials. In the first stage, filtering techniques are used to remove high-frequency or low-frequency noise interference, making the basic signal outline clearer. Then, in the second stage, amplitude adjustments such as signal amplification or attenuation are performed to bring the signal into a suitable numerical range for subsequent processing. After this series of preprocessing steps, the generated standard electrical signal data has higher quality, providing a solid and reliable data foundation for subsequent accurate calibration.

[0053] The design of the dynamic evaluation and fine calibration process greatly enhances the targeting of calibration. Traditional calibration methods often adopt a "one-size-fits-all" approach, performing the same level of calibration regardless of the actual operating status of the electricity metering device. However, the dynamic evaluation step in this invention, by carefully comparing the characteristics of the measured signal with those of the ideal signal, can keenly capture the current operating status of the electricity metering device. When the calibration status indicator indicates that fine calibration is required, the standard electrical signal data is adaptively divided into multiple signal segments, and each signal segment is independently verified. This approach is similar to a doctor conducting a detailed specialized examination of a patient, accurately locating the problematic signal segment. Based on the verification results, qualified signal segments are filtered out, avoiding over-calibration of qualified parts and focusing efforts on addressing problematic parts. This significantly improves calibration efficiency and accuracy, reduces unnecessary calibration operations, and saves time and resources.

[0054] Segmenting, sorting, and performing correlation analysis on electrical parameter values ​​are crucial steps in gaining a deeper understanding of the operational status of electricity metering devices. After extracting all electrical parameter values ​​from the remaining filtered signal segments, the deviation of each parameter value from its corresponding reference value is calculated. Based on the positive or negative characteristics of the deviation, the electrical parameter values ​​are divided into normal parameter clusters and abnormal parameter clusters. This segmentation method distinguishes electrical parameters of different natures, much like classifying and storing different categories of items, facilitating subsequent analysis and processing.

[0055] Local sorting operations are performed on electrical parameter values ​​in normal and abnormal parameter clusters to form ordered parameter sequences. Adjacent electrical parameter values ​​in these sequences are then paired as parameter groups to be verified, further uncovering the potential relationships between electrical parameters. A pre-built parameter correlation model is used to perform correlation analysis on each parameter group to be verified, outputting a correlation score. This approach provides a comprehensive understanding of the degree of mutual influence between various electrical parameters. For example, in an energy metering device, there may be a specific correlation between current and voltage parameters; correlation analysis can reveal this correlation, thus more accurately assessing the operating status of the energy metering device. The number of valid correlation groups with scores exceeding a threshold is counted, and combined with electrical parameter values ​​that were not paired, an overall calibration index is calculated. This comprehensive analytical method makes calibration decisions more scientific and reasonable, moving beyond relying solely on single parameter judgments to considering the interrelationships of multiple parameters and the overall situation to make more realistic calibration decisions. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the working principle of the precise calibration method for electricity metering described in this invention.

[0057] Figure 2This is a flowchart of the raw electrical signal acquisition and multi-stage signal preprocessing.

[0058] Figure 3 This is a flowchart for extracting electrical parameter values, calculating deviations, and dividing clusters. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 This invention provides a precise calibration system and method for electricity metering. The method includes: acquiring the raw electrical signal of the electricity metering device to be calibrated, and performing multi-stage signal preprocessing on the raw electrical signal to generate standard electrical signal data; the multi-stage signal preprocessing involves noise cancellation, amplitude normalization, and interference removal. The standard electrical signal data is dynamically evaluated by comparing the measured signal characteristics with the ideal signal characteristics to generate a calibration status identifier; when the calibration status identifier indicates that fine calibration is required, the standard electrical signal data is adaptively segmented into multiple electrical signal segments, and each electrical signal segment is independently verified, filtering out qualified electrical signal segments based on the verification results. All electrical parameter values ​​are extracted from the remaining filtered electrical signal segments, and the deviation value of each electrical parameter value from the corresponding reference parameter value is calculated. Based on the positive and negative characteristics of the deviation value, the electrical parameter values ​​are divided into normal parameter clusters and abnormal parameter clusters. A local sorting operation is performed on the electrical parameter values ​​in the normal parameter clusters and abnormal parameter clusters to form an ordered parameter sequence, and adjacent electrical parameter values ​​in the sequence are paired as parameter groups to be verified. A pre-built parameter correlation model is used to perform correlation analysis on each parameter group to be verified, outputting a correlation score and counting the number of valid correlation groups with scores exceeding a threshold. Combining the number of valid correlation groups and the electrical parameter values ​​that did not participate in pairing, an overall calibration index is calculated. Based on a comparison of the overall calibration index with the dynamic calibration threshold, it is determined whether to initiate the calibration process.

[0061] Example 1: See Figure 2A high-precision sensor captures the analog voltage and current waveforms of the energy metering device to be calibrated in real time at a sampling frequency of 10,000 times per second. The analog signal is converted into the original digital electrical signal by a 24-bit analog-to-digital converter. The wavelet transform algorithm uses the Daubechies-8 wavelet basis to perform a 5-level multi-resolution decomposition on the original electrical signal, separating the random noise in the high-frequency components of the signal from the fundamental component. The noise cancellation process sets the detail coefficients to zero through threshold processing, and the reconstructed signal yields a preliminary purified signal with improved signal-to-noise ratio. The amplitude normalization process uses the minimum-maximum scaling technique to linearly map the peak-to-peak value of the preliminary purified signal to the standard range of [-1,1]. The digital filter is designed as an 8th-order Butterworth low-pass filter with a cutoff frequency set to 15 times the fundamental frequency of the signal, effectively suppressing high-frequency interference components while maintaining the signal phase characteristics. The signal quality assessment unit is integrated into the digital signal processor. The signal-to-noise ratio (SNR) is calculated using the power spectral density estimation method, converting the ratio of useful power to noise power within the signal band into decibels (dB). The preset threshold is set to 30 dB. The stability index is obtained by calculating the reciprocal of the signal variance within a sliding window, requiring that the stability index fluctuation range of 100 consecutive windows does not exceed 5%. When the SNR is higher than the threshold and the stability index meets the fluctuation requirements, the system marks the processed signal as standard electrical signal data and stores it in the buffer memory. When any indicator fails to meet the standard, the acquisition parameter adjustment module gradually increases the sampling rate to 20,000 times / second or decreases it to 5,000 times / second, while adjusting the sensor gain coefficient and restarting the signal acquisition process until the standard is met.

[0062] The key signal feature point extraction module locates local extrema in standard electrical signal data as peak features, calculates the mean feature within the window using an arithmetic mean, and extracts instantaneous phase information through Hilbert transform. Specifically, this is achieved using a finite impulse response filter, with filter coefficients designed based on the impulse response of an ideal Hilbert transform. Convolution operations are performed in a digital signal processor, generating an analytic signal from which instantaneous phase information is extracted. The analytic signal consists of the original signal as the real part and the transformed signal as the imaginary part. The instantaneous phase is obtained by calculating the arctangent values ​​of the imaginary and real parts in the four quadrants. The phase information is stored in radians for subsequent point-by-point matching with ideal signal feature points in the database. Ideal signal feature points are stored in a relational database, containing theoretical peak, mean, and phase data under a standard frequency of 50Hz. The point-by-point matching process calculates the Euclidean distance between the measured feature points and the ideal feature points, with a tolerance range set to ±2% of the absolute value of the feature value according to the IEC electrical energy metering standard. When all feature point errors are less than the lower tolerance limit, the system generates a calibration pass flag in binary format and terminates the process; when all feature point errors exceed the upper tolerance limit, a calibration failure flag is generated and an abnormal alarm is triggered; when some feature point errors are within the tolerance range while others exceed it, a fine calibration flag will activate the subsequent signal segmentation process. The high-precision sensor uses a temperature-compensated current and voltage transformer, and the sampling timing is controlled by a precision crystal oscillator. Quantization errors during analog-to-digital conversion are dispersed using dithering technology. The threshold selection for wavelet transform uses a rigrsure adaptive rule to ensure that noise of different intensities is suppressed. Amplitude normalization processing introduces an overflow protection mechanism to prevent data truncation caused by transient overvoltages. The group delay error of the digital filter is eliminated through forward and reverse filtering to avoid signal distortion. The signal quality assessment unit monitors signal characteristics in real time. The signal-to-noise ratio is calculated using the periodogram method, and the stability index assessment introduces a time-weighted factor, giving higher weight to recent data.

[0063] The feature point extraction algorithm employs multi-scale analysis, peak detection combines first-order difference and threshold comparison, and mean calculation uses a weighted average method to reduce edge effects. Phase information is verified through zero-crossing point detection. Ideal feature points in the database are dynamically updated according to standard power grid operating conditions, and the matching process uses a kd-tree to accelerate nearest neighbor search. Tolerance range settings consider the impact of temperature drift and implement a dynamic compensation strategy; the identifier generation logic is implemented using a state machine, with three identifiers corresponding to different program jump addresses. The signal preprocessing stage and the dynamic evaluation stage form a pipelined operation, with a double-buffering mechanism for seamless connection between the output buffer of the previous stage and the input buffer of the next stage. The re-acquisition process includes a maximum retry limit to avoid infinite loops; feature point matching results are persistently stored for generating calibration certificates. The entire implementation process embodies the synergy between hardware acquisition and software processing. High-precision sensors provide the raw data foundation, digital signal processing algorithms enable refined analysis, and evaluation results directly guide calibration decisions.

[0064] Example 2: See Figure 3 The sliding window algorithm employs a dynamic window management mechanism. The window width is adaptively adjusted based on the fundamental frequency of the standard electrical signal data. When a 50Hz fundamental frequency is detected, the window width is set to 100 milliseconds; for a 60Hz system, it is adjusted to 83.3 milliseconds. Each electrical signal segment contains a fixed 1024 sampling points, and waveform continuity is ensured through a 50% overlap rate. The window sliding step size is calculated as half the window width to achieve complete signal coverage. The consistency score of the electrical signal segment is calculated using the variance coefficient and autocorrelation function of the sampling points within the segment. The variance coefficient reflects the segment's stability, while the autocorrelation function detects the periodicity of the waveform. A passing threshold of 0.95 points is set; segments with scores higher than this threshold are considered passing and immediately removed from the processing queue. The feature extraction algorithm employs an improved Fast Fourier Transform to analyze each remaining electrical signal segment. The effective voltage value is obtained by calculating the root mean square value of the sampling points within the segment. Current harmonic content analysis covers harmonic components from the 2nd to the 13th order and uses total harmonic distortion (THD) quantification. The power factor is calculated using the cosine of the phase difference between the voltage and current segments. A standard parameter table is stored in an embedded database. Reference parameter values ​​are set according to the IEC 61000-4-30 standard, including a 230V rated voltage, a 5% THD limit, and a 0.98 reference power factor. Absolute deviation is calculated using the algebraic difference between the reference parameter value and the measured value. Positive or zero deviations are categorized into the normal parameter cluster, and the deviation value is recorded. Negative deviations are categorized into the abnormal parameter cluster, and the negative offset value is saved. Parameter subset division is based on the timestamp sequence of the electrical signal segments, grouping parameter values ​​within the continuous time domain into the same subset.

[0065] The dynamic adjustment of the sliding window algorithm is achieved through real-time frequency monitoring by a digital signal processor. Fundamental frequency detection uses a zero-crossing detection algorithm in conjunction with a digital phase-locked loop (PLL). The window width is inversely proportional to the frequency to ensure each window contains an integer multiple of the complete cycle. The overlapping design of electrical signal segments employs a circular buffer management system. When a new segment is generated, 512 sampling points following the previous segment are retained. Segment marking uses both timestamps and sequence numbers. The consistency score calculation module integrates a dedicated hardware accelerator. Variance coefficient calculation uses the Welford online algorithm to avoid floating-point overflow. Specifically, the variance coefficient calculation is implemented through a dedicated hardware accelerator. This algorithm updates the mean and squared difference point-by-point using an iterative approach, avoiding the floating-point overflow problem caused by large number subtraction in traditional methods. The hardware accelerator design includes multiple pipeline stages, each processing one sampling point. Initially, the mean and squared difference are set to zero. As sampling points are input, the algorithm dynamically updates these statistics, using accumulators and shift registers to store intermediate results, ensuring numerical stability. The autocorrelation function calculation employs an accelerated FFT method; the acceptable threshold is dynamically fine-tuned based on the on-site operating environment, increasing by 0.01 for every 10 degrees Celsius increase in ambient temperature. The Fast Fourier Transform (FFT) in the feature extraction process uses a radix-2 algorithm with window function compensation; waveform factor correction is introduced in the calculation of the effective voltage value; and windowed interpolation FFT is used for current harmonic analysis to reduce spectral leakage. The standard parameter table is synchronized with the cloud-based standard library via HTTPS, and the benchmark parameter values ​​have a version management mechanism to support rollback operations. A temperature compensation coefficient is added to the calculation of absolute deviation values; the data structure for normal parameter clusters uses a red-black tree for fast retrieval, while abnormal parameter clusters are stored using a time-series database. The parameter subset partitioning algorithm is based on density clustering principles, automatically grouping electrical parameter values ​​with time intervals less than 200 milliseconds, and establishing an independent index for each subset to support fast access.

[0066] The segmentation of electrical signal fragments employs a parallel pipeline architecture, with four processing cores simultaneously executing window sliding operations. Fragment data is transferred via DMA to reduce CPU load. Consistency score evaluation incorporates a machine learning model to assist in judgment, using waveform features from historical fragments to train a support vector machine classifier to improve accuracy. The feature extraction stage is equipped with a hardware coprocessor specifically for FFT operations and harmonic analysis, and power factor calculation incorporates a dual-path verification mechanism of instantaneous and average values. A lag interval is set for the division of normal and abnormal parameter clusters to avoid frequent parameter switching near thresholds, and the size of parameter subsets is limited to 100 parameter values ​​to prevent memory overflow. A mirror expansion method is used for sliding window boundary processing to avoid edge effects, and a Hanning window function is applied to each electrical signal fragment to reduce spectral leakage. Detailed logs are recorded for the filtering of qualified fragments for auditing purposes, and remaining fragments are automatically compressed and stored to save storage space. Feature extraction results undergo multiple verification logics, and outliers automatically trigger a re-acquisition process. Data exchange between normal and abnormal parameter clusters uses a transaction mechanism to ensure consistency, and a topological relationship graph is established between parameter subsets to support cross-subset analysis. The entire implementation process embodies the deep integration of signal processing technology and database management. The sliding window algorithm provides the foundation for data segmentation, the feature extraction algorithm realizes parameter quantization, and cluster partitioning constructs a data analysis framework.

[0067] Example 3: In the parameter sorting and pairing stage of the precise calibration method for electricity metering, the local sorting operation is performed on the electrical parameter values ​​in the normal parameter cluster and the abnormal parameter cluster. These electrical parameter values ​​are derived from parameters such as the effective voltage value, current harmonic content, and power factor extracted from the remaining electrical signal segments after filtering. Each electrical parameter value carries a timestamp and segment identification information. The normal parameter cluster contains electrical parameter values ​​with positive or zero absolute deviation values, and the abnormal parameter cluster contains electrical parameter values ​​with negative absolute deviation values. Both clusters are divided into multiple parameter subsets according to the source of the electrical signal segments, and the electrical parameter values ​​within the subsets are stored in an array structure. The sorting key value calculation uses a symbol-weighted formula, expressed as:

[0068]

[0069] Where: symbol This represents the sorting key value, with the same units as the electrical parameter value, and the symbol... Represents the symbolic factor, symbol Represents the raw numerical values ​​of the electrical parameters. Within each subset of parameters in a normal parameter cluster, the electrical parameter values ​​are sorted by key value. Ascending order is used to generate a normal ordered sequence. The sorting algorithm uses in-place partitioning quicksort, and the comparison function is designed to determine the sign of the key-value difference. Within each subset of parameters in the abnormal parameter cluster, the electrical parameter values ​​are sorted by key value. The sequence is sorted in ascending order. Since the sign factor is negative, ascending key values ​​are equivalent to descending original values, resulting in an abnormally ordered sequence. After sequence generation, two adjacent electrical parameter values ​​in the normally ordered sequence are automatically combined into a normal parameter group to be verified, while two adjacent electrical parameter values ​​in the abnormally ordered sequence are combined into an abnormal parameter group to be verified. Pairing is achieved by traversing the sequence index, and each group of parameter values ​​is encapsulated as a structure containing key-value pairs. For isolated electrical parameter values ​​that do not belong to any parameter subset, these values ​​are not included in the subset due to discontinuous timestamps or missing source fragments, and are temporarily not involved in sorting and pairing, but are kept in an independent buffer and marked as unassociated parameters.

[0070] When electrical parameter values ​​are loaded into the sorting module, memory is allocated as a contiguous address space, and each parameter value is accompanied by metadata including parameter type and deviation; sign factor. The assignment is done by querying the parameter cluster type register. The sign factor of a normal parameter cluster is hardcoded to 1, and the sign factor of an abnormal parameter cluster is hardcoded to -1. Sort key value The calculations are performed by arithmetic logic units using single-cycle multiplication operations, and key-value pairs are stored in floating-point format to support high-precision parameter values. The generation of normal ordered sequences uses a recursive implementation of the quicksort algorithm. The pivot value is selected using the median of three numbers to avoid worst-case scenarios, and ascending order is controlled by a comparator output for data exchange. The sorting process for abnormally ordered sequences is similar, but the sign factor is reversed during key-value calculation, and the sequence traversal direction is consistent with that of normal sequences. In the pairing phase, the index of the normal ordered sequence starts from 0 and increments by 2. The electrical parameter values ​​at indices i and i+1 are packaged into normal parameter groups to be verified. The pairing logic for abnormally ordered sequences is the same, and the number of groups generated depends on the integer part of the sequence length divided by 2. Unassociated parameters are stored separately in a linked list, equipped with a timestamp index for subsequent retrieval.

[0071] The data flow for sorting operations is managed by a direct memory access controller. Electrical parameter values ​​are read in batches from the cluster memory into the cache. Pipeline operations between the sorting key-value calculation unit and the sorting engine reduce latency. The subset size of normal parameter clusters changes dynamically, and the sorting algorithm adaptively adjusts the recursion depth. After sequence generation, the address pointer is stored in a queue. For abnormal parameter clusters, a stability protection mechanism is introduced: electrical parameter values ​​with the same key value maintain their original relative order, and a circular buffer is used for pairing group allocation to avoid memory fragmentation. Isolation of unassociated parameters is based on a subset mapping table. An outlier detection algorithm calculates a time interval threshold; electrical parameter values ​​with intervals exceeding 200 milliseconds are marked as outliers. (Sign factor) The configuration is set via hardware pin levels; a high level corresponds to normal parameter clusters, and a low level corresponds to abnormal parameter clusters. The key-value calculation circuit integrates a multiplier unit. Sorting key values... The floating-point representation conforms to the IEEE 754 standard, and comparison operations use precision tolerance to prevent misjudgments due to floating-point errors. Ascending order of normally ordered sequences ensures that smaller parameters are paired first, while ascending order of key values ​​in abnormally ordered sequences prioritizes pairing of larger original parameters. The continuous index design of pairing groups ensures locality of reference and optimizes cache hits. Unassociated parameters are handled using a lazy approach, only re-checking subset ownership when new data arrives. The entire sorting and pairing process reflects a balance between computational efficiency and data structure. The symbolic weighting formula unifies the sorting logic, adjacent pairings simplify group management, and isolated values ​​reduce complexity. Implementation details encompass the collaboration between hardware acceleration and software coroutines, optimizing the time complexity of the sorting algorithm, maintaining constant space complexity for pairing operations, and improving system throughput through asynchronous processing of unassociated parameters.

[0072] Preprocessing of electrical parameter values ​​before sorting includes range checking and outlier filtering. Parameter values ​​exceeding their physical range are automatically truncated to a reasonable range, and overflow protection is added to the sorting key value calculation. Within the parameter subset of the normal parameter cluster, electrical parameter values ​​are grouped and sorted by type: voltage parameters are serialized separately, current harmonic parameters are serialized separately, and power factor parameters are serialized separately to prevent comparison errors between different parameter types. Negative sorting key values ​​for outlier parameter clusters are handled using two's complement representation, and the comparator design supports signed integer arithmetic. After sequence generation, electrical parameter values ​​of the normal ordered sequence are stored in a linear array in index order, while the storage structure for outlier ordered sequences is identical. Pairing operations traverse the array once to complete group construction. The buffer size for unassociated parameters is dynamically adjusted, and the memory allocation strategy is based on the least recently used algorithm. (Sign factor) The assignment logic is embedded in the sorting controller firmware. Cluster type identification is determined by the sign bit of the deviation value, and key-value calculation is pipelined to improve throughput. The worst-case time complexity of the sorting algorithm is O(n), space complexity is O(n), and pairing operation time complexity is O(n). The retrieval interface for unassociated parameters provides query functionality by time range. The implementation process integrates an error detection mechanism, saving checkpoints when sorting is interrupted, and pairing group checksums verify data integrity. The sorting key-value pairs for electrical parameters are... After calculation, the original value retains its read-only attribute, and the serialization process does not modify the source data.

[0073] Ascending order of normal ordered sequences facilitates pattern recognition in subsequent correlation analysis, while descending order of abnormal ordered sequences highlights anomalous trends. The structure of paired groups includes pointers to the original parameter values. Unassociated parameters trigger re-subsetting attempts when the system is idle. The core role of the sorting key-value formula is emphasized; the formula uniformly handles normal and abnormal clusters, the sorting results drive the pairing process, and isolated value handling ensures system robustness. Technical implementation combines digital circuits and algorithm optimization, with hardware-based formula calculation, software scheduling for sorting and pairing, and background management of unassociated parameters, forming an efficient parameter organization framework.

[0074] Example 4: In the parameter association model construction and analysis stage of the precise calibration method for electricity metering, the pre-constructed parameter association model adopts a graph neural network architecture. The model training data comes from a historical calibration task record library. These records contain electrical parameter values, benchmark parameter values, and deviation data fields. The data collection time span covers a continuous 24-month operating cycle of the electricity metering device. During the construction of the training dataset, the original historical records are cleaned and labeled. After removing obvious outliers, they are randomly divided into training subsets and test subsets. The training subset accounts for 70% for model parameter learning, and the test subset accounts for 30% for model performance verification. The graph neural network initialization settings include an input layer, three hidden layers, and an output layer. The number of nodes in the input layer corresponds to the electrical parameter feature dimension. The number of nodes in the hidden layer are 128, 64, and 32 respectively. The output layer uses the sigmoid activation function to generate association probability values.

[0075] Referring to Table 1, the model training uses an adaptive moment estimation algorithm for backpropagation optimization. Training subset data is input into the network in batches, with each batch containing 32 samples. The loss function uses binary cross-entropy to calculate the difference between the predicted value and the true label. After each iteration, a test subset is used to evaluate the model's association prediction accuracy. The accuracy is calculated as the ratio of correctly predicted samples to the total number of samples. When the accuracy improvement after 10 consecutive iterations is less than 0.001, the learning rate automatically decays to 0.5 times its original value. The training termination condition is set to accuracy convergence or the number of iterations reaching 1000. The model parameters are then stored as a binary file for use in the inference phase. When the parameter groups to be verified are input into the model, both normal and abnormal parameter groups undergo forward propagation calculations. Each group outputs an association probability value between 0 and 1. Groups with a probability value greater than 0.7 are considered valid association groups, and the number of valid association groups is accumulated in real-time using a hardware counter.

[0076] Table 1: Parameter Configuration Table for Graph Neural Network Layer Structure

[0077]

[0078] The historical calibration task record library is stored using a time-series database structure. Each record includes a timestamp, electrical parameter type, numerical value, and deviation flag fields. Data cleaning rules include range checks and consistency verification. The training and test subsets are divided using stratified sampling to ensure consistent distribution ratios for different parameter types. The graph neural network node design supports edge feature learning, and the relationships between electrical parameter values ​​are modeled through a message passing mechanism. The gradient clipping threshold during model training is set to 1.0 to prevent gradient explosion, the initial learning rate is 0.01, and cosine annealing is used as the decay strategy to enhance convergence stability. Data preprocessing for the parameter groups to be verified includes numerical standardization and missing value imputation. The standardization method uses Z-score transformation to map parameter values ​​to a zero-mean, unit-variance distribution, and missing values ​​are imputed using the median of the same parameter type. Forward propagation computation is accelerated using parallel processing of a graphics processor. Each parameter group to be verified is constructed as a fully connected subgraph, and node features include parameter value, parameter type, and time interval. The output precision of the association probability value is retained to four decimal places, and the determination of valid association groups uses comparator circuitry for hardware-level acceleration. The statistical results of the number of effective associated groups are displayed in real time on the monitoring interface and written to the system log for subsequent analysis.

[0079] The model training environment is configured with a dual-GPU workstation and 128GB of memory. Training data loading utilizes an NVMe SSD to improve I / O throughput. The loss curve during training is displayed in real-time on the monitoring terminal. An early stopping mechanism forcibly terminates training when the validation set loss fails to decrease for five consecutive iterations. The model inference phase is deployed on an embedded system, and the graph neural network model is converted to the TensorRT engine to improve inference speed. Batch processing of parameter groups to be validated employs a pipelined architecture; while one set of parameters is being calculated, the next set is preprocessed, achieving a system throughput of 1000 sets of parameters per second. Feature engineering of electrical parameters includes derived feature generation, deriving statistical features such as rate of change, acceleration, and fluctuation coefficients from the original parameter values. The adjacency matrix construction of the graph neural network is based on the physical relationships between parameters; strong connections are established between voltage and current parameters, and weak connections are established between parameters of the same type. Regularization techniques during training include L2 weight decay and label smoothing to prevent overfitting of the training data. Before the statistical results of the number of effective association groups are included in the calibration index calculation, a majority voting mechanism is used to filter out random associations; each group is counted only if it is confirmed as a valid association group by at least three independent models. Model version management employs a Git-like system, generating a unique hash identifier for each training iteration, supporting model rollback and A / B testing. The correlation analysis results of the parameter groups to be verified generate a visual graph, with node size representing parameter values ​​and edge thickness representing correlation strength. System runtime resource monitoring includes GPU utilization and memory usage, automatically triggering model sharding when thresholds are exceeded. The entire parameter correlation model implementation process demonstrates a deep integration of machine learning and the field of electricity metering. Graph neural network structures capture complex relationships between parameters, adaptive training mechanisms ensure model generalization ability, and hardware acceleration technology meets real-time requirements.

[0080] Example 5: In the calibration index calculation and decision-making process of the precise calibration method for electricity metering, the calculation of the overall calibration index requires integrating the number of effective correlation groups and the information of electrical parameter values ​​that have not participated in pairing. The number of effective correlation groups comes from the analysis results of the parameter correlation model on the normal and abnormal parameter groups to be verified. The electrical parameter values ​​that have not participated in pairing refer to those residual parameters that failed to be included in the parameter subset due to isolation in the local sorting operation. The calculation process first applies a natural logarithmic transformation to the number of effective correlation groups to compress the data scale. The transformation result is multiplied by the total deviation range of the normal parameter cluster, which is obtained by calculating the difference between the maximum and minimum values ​​of the deviation of all electrical parameter values ​​in the cluster. Then, the above product is multiplied by the total deviation range of the abnormal parameter cluster, which is calculated using the same method, finally obtaining the correlation contribution component. The residual deviation of each electrical parameter value that has not participated in pairing is calculated with its corresponding reference parameter value. The residual deviation is processed by taking the absolute value to avoid sign cancellation. All residual deviations are summed to obtain the non-correlated contribution component. The associated and unassociated contribution components are added in the algebraic space, and the result is multiplied by a weighting factor. The weighting factor is obtained by looking up a table according to the type of electricity metering device. The weighting factor is 0.9 for residential electricity metering devices and 1.1 for industrial electricity metering devices. Finally, the overall calibration index value is derived.

[0081] The dynamic calibration threshold adjustment mechanism monitors the output signals of the ambient temperature sensor and the load current transformer in real time. For every 1 degree Celsius change in ambient temperature, the calibration threshold is adjusted by 0.5 reference units; for every 10 amperes increase in workload, the calibration threshold is adjusted by 1.2 reference units. The calibration decision module compares the overall calibration index with the dynamic calibration threshold. When the overall calibration index is greater than or equal to the dynamic calibration threshold, the digital comparator outputs a high level to trigger a calibration command. This command is sent to the calibration circuit of the energy metering device via the fieldbus, initiating an automatic calibration mechanism to reconfigure the gain and offset parameters of the metering chip. When the overall calibration index is less than the dynamic calibration threshold, the comparator outputs a low level to generate a sustaining state command. This command keeps the system in its current operating mode, skips the calibration process, and records the decision in the log.

[0082] In a specific example, an industrial power metering device has 15 effective associated groups. The normal parameter cluster includes voltage values ​​of 230.1V, 230.3V, and 230.0V, with deviations of +0.1V, +0.3V, and +0.0V respectively, for a total deviation range of 0.3V. The abnormal parameter cluster includes power factor values ​​of 0.85, 0.82, and 0.80, with deviations of -0.13, -0.16, and -0.18 respectively, for a total deviation range of 0.05. Unpaired electrical parameter values ​​include an isolated current harmonic parameter value of 4.8%, a reference parameter value of 5.0%, and a residual deviation of 0.2%. When calculating the associated contribution component, the natural logarithm of the effective associated group number 15 is 2.708, multiplied by the total deviation range of the normal parameter cluster (0.3V) to obtain 0.8124, and then multiplied by the total deviation range of the abnormal parameter cluster (0.05) to obtain 0.04062. The non-associated contribution component is directly calculated using the residual deviation of 0.2%. The weighting factor for industrial-type electricity metering devices is 1.1, and the overall calibration index is calculated as follows: The ambient temperature sensor displays a current temperature of 45 degrees Celsius. At a reference temperature of 25 degrees Celsius, the calibration threshold is 0.04. A temperature deviation of 20 degrees Celsius corresponds to a threshold adjustment of 10 reference units, resulting in an actual threshold of 0.05. The load current transformer measures 150 amperes. At a reference load of 100 amperes, the threshold is adjusted by 6 reference units, resulting in a final dynamic calibration threshold of 0.056. Since the overall calibration index of 0.04468 is less than the dynamic calibration threshold of 0.056, the system generates a maintenance command, and the energy metering device continues its current operating state.

[0083] In the overall calibration index calculation process, the natural logarithmic transformation is implemented on an FPGA using the CORDIC algorithm. The total deviation range calculation for normal parameter clusters employs a hardware extreme value detection circuit, while the total deviation range calculation for abnormal parameter clusters shares the same hardware resources. The residual deviation calculation for unpaired electrical parameter values ​​is completed by a dedicated arithmetic logic unit, and the accumulation operation uses a pipelined structure to improve throughput. The weighting factor lookup table is stored in EEPROM memory, and the address bus automatically addresses based on the model code of the energy metering device. The ambient temperature compensation coefficient for the dynamic calibration threshold is stored in a lookup table, temperature sensor data is read via the I2C bus, and load measurement values ​​are digitally filtered after ADC conversion. The calibration decision comparator adopts a window comparator design to prevent oscillation near the threshold. The natural logarithmic transformation of the number of effective association groups is accelerated using a lookup table method, with the logarithm table pre-stored in ROM, and the input value range covering 0-1000 groups. When calculating the total deviation range of normal parameter clusters, the maximum value detection circuit compares the voltage values ​​of all deviations, and the minimum value detection circuit works synchronously, with the difference calculation completed in a single clock cycle. The calculation process for the total deviation range of abnormal parameter clusters is the same, and the calculation results for both clusters are temporarily stored in a register file. The residual deviation calculation for electrical parameter values ​​not involved in pairing adds sign judgment logic, and the absolute value is then fed into an addition tree structure. Weighting factor multiplication uses a hardware multiplier, supporting floating-point arithmetic. The adjustment amount for the dynamic calibration threshold is calculated using piecewise linear interpolation; temperature compensation and load compensation are calculated separately and then superimposed onto the reference threshold. After the calibration command is generated, calibration parameters are sent via the SPI interface, and a status maintenance command is written to the system status register.

[0084] In the specific examples, all numerical calculations use fixed-point arithmetic to ensure determinism. The logarithmic transformation of the 15 effective correlation groups is obtained as Q-format fixed-point numbers through table lookup. The deviation of normal parameter clusters is stored as 12-bit fixed-point numbers, and the accuracy of the total deviation range calculation reaches 0.01V. The deviation of abnormal parameter clusters is represented as 16-bit fixed-point numbers, and the total deviation range calculation retains four decimal places. The residual deviation calculation of electrical parameter values ​​that did not participate in pairing uses the same numerical format. The weighting factor multiplication operation is rounded to avoid error accumulation. The adjustment amount of the dynamic calibration threshold is calculated using saturation arithmetic to prevent overflow, and the reference units for temperature compensation and load compensation are converted to the same dimension. The comparator design includes hysteresis characteristics, and the comparison result between the overall calibration index and the dynamic calibration threshold is latched for one working cycle.

[0085] The overall calibration index calculation circuit adopts a three-stage pipeline architecture. The first stage calculates the associated contribution components, the second stage calculates the non-associated contribution components, and the third stage performs weighted summation. The dynamic calibration threshold generation module runs independently, updating the threshold value every second. The calibration decision logic performs a comparison operation once per power cycle to ensure real-time response. In the specific example, the serial number of the industrial energy metering device triggers the weight factor query address, the temperature sensor is installed inside the metering box, and the load current transformer uses a Rogowski coil structure. All calculated parameters during system operation are recorded in non-volatile memory for post-audit analysis. During the overall calibration index calculation, the update of the number of effective association groups is synchronized with the parameter association model analysis, and the total deviation range of normal and abnormal parameter clusters is recalculated with each parameter update. The number of electrical parameter values ​​not involved in pairing changes dynamically, and the residual deviation summation circuit supports dynamic input. The weight factor is updated periodically based on the annual inspection records of the energy metering device, and the benchmark value of the dynamic calibration threshold can be configured remotely. The calibration command contains specific calibration parameter values, and the maintenance status command records the system timestamp. The entire implementation method embodies the close coupling between index calculation and threshold adaptation, the numerical processing takes into account both accuracy and real-time performance, and the decision logic ensures the accurate triggering of calibration actions.

[0086] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for accurate calibration based on electric energy metering, characterized in that, The method includes: The raw electrical signals of the power metering device to be calibrated are acquired, and the raw electrical signals are subjected to multi-stage signal preprocessing to generate standard electrical signal data. The standard electrical signal data is dynamically evaluated. By comparing the measured signal characteristics with the ideal signal characteristics, a calibration status identifier is generated. When the calibration status identifier indicates that fine calibration is required, the standard electrical signal data is adaptively divided into multiple electrical signal segments, and each electrical signal segment is independently verified. Based on the verification results, qualified electrical signal segments are filtered out. Extract all electrical parameter values ​​from the remaining electrical signal segments after filtering, and calculate the deviation of each electrical parameter value from the corresponding reference parameter value. Based on the positive and negative characteristics of the deviation value, divide the electrical parameter values ​​into normal parameter clusters and abnormal parameter clusters. A local sorting operation is performed on the electrical parameter values ​​in the normal parameter cluster and the abnormal parameter cluster to form an ordered parameter sequence, and adjacent electrical parameter values ​​in the sequence are paired as a parameter group to be verified. A pre-built parameter association model is used to perform correlation analysis on each parameter group to be verified, output the association score, and count the number of valid association groups with scores exceeding the threshold. Based on the number of effective associated groups and the electrical parameter values ​​of those not involved in pairing, the overall calibration index is calculated, and the calibration process is determined to be initiated based on the comparison between the overall calibration index and the dynamic calibration threshold. When calculating the overall calibration index by combining the number of effective associated groups and the electrical parameter values ​​that did not participate in pairing, the following is included: Calculate the logarithmic transformation value of the number of effective association groups, and multiply it by the total deviation range of the normal parameter cluster and the total deviation range of the abnormal parameter cluster to obtain the association contribution component; For electrical parameter values ​​that are not paired, calculate their residual deviations from the corresponding reference parameter values, and sum them to obtain the non-associated contribution components; The overall calibration index is derived by summing the associated and unassociated contribution components and multiplying them by a weighting factor determined according to the type of metering device.

2. The method for precise calibration based on electricity metering according to claim 1, characterized in that, When acquiring the raw electrical signal from the energy metering device to be calibrated, and performing multi-stage signal preprocessing on the raw electrical signal to generate standard electrical signal data, the process includes: The original electrical signal is acquired in real time by a high-precision sensor, and the noise of the original electrical signal is eliminated by applying a wavelet transform algorithm to obtain a preliminary purified signal. The preliminary purified signal is subjected to amplitude normalization processing to scale the signal amplitude to a standard range, and a digital filter is used to remove high-frequency interference components. The processed signal is input into the signal quality evaluation unit to calculate the signal-to-noise ratio and stability index. If the signal-to-noise ratio is higher than the preset threshold and the stability index meets the requirements, the standard electrical signal data is output. If the signal-to-noise ratio or stability index does not meet the standard, adjust the acquisition parameters and reacquire the original electrical signal.

3. The method for precise calibration based on electricity metering according to claim 2, characterized in that, When dynamically evaluating the standard electrical signal data and generating a calibration status identifier by comparing the measured signal characteristics with the ideal signal characteristics, the process includes: Key signal feature points, including peak value, mean value and phase information, are extracted from the standard electrical signal data and matched point by point with ideal signal feature points in the database. A calibration pass flag is generated when the errors between all measured signal feature points and ideal signal feature points are below the tolerance range. A calibration failure flag is generated when the errors between all measured signal feature points and ideal signal feature points exceed the tolerance range. When the error of some measured signal feature points is within the tolerance range but partially exceeds it, a fine calibration mark is generated.

4. The method for precise calibration based on electricity metering according to claim 3, characterized in that, When adaptively segmenting the standard electrical signal data into multiple electrical signal segments, independently verifying each electrical signal segment, and filtering out qualified electrical signal segments based on the verification results, the process includes: Based on the characteristics of the signal time series, the standard electrical signal data is divided into overlapping electrical signal segments using a sliding window algorithm, with each segment containing a fixed number of sampling points; For each electrical signal segment, calculate the consistency score of the signal characteristics within the segment. If the consistency score is higher than the qualified threshold, the electrical signal segment is deemed qualified and filtered out, and only electrical signal segments with a consistency score lower than the qualified threshold are retained.

5. The method for precise calibration based on electricity metering according to claim 4, characterized in that, When extracting all electrical parameter values ​​from the remaining electrical signal segments after filtering, calculating the deviation of each electrical parameter value from the corresponding reference parameter value, and classifying the electrical parameter values ​​into normal parameter clusters and abnormal parameter clusters based on the positive or negative characteristics of the deviation, the process includes: A feature extraction algorithm is used to extract values ​​from each remaining electrical signal segment, including the effective voltage value, current harmonic content, and power factor; For each electrical parameter value, the corresponding reference parameter value is obtained by querying the standard parameter table, and the absolute deviation value is calculated. If the absolute deviation value is positive or zero, the electrical parameter value is classified into the normal parameter cluster and the positive deviation amount is recorded. If the absolute deviation value is negative, the electrical parameter value is classified into the abnormal parameter cluster and the negative deviation amount is recorded. The normal parameter cluster and the abnormal parameter cluster are subgrouped according to the source of the electrical signal segment, forming multiple parameter subsets.

6. The method for precise calibration based on electricity metering according to claim 5, characterized in that, When performing a local sorting operation on the electrical parameter values ​​in the normal parameter cluster and the abnormal parameter cluster to form an ordered parameter sequence, and pairing adjacent electrical parameter values ​​in the sequence into a parameter group to be verified, the process includes: Arrange the electrical parameter values ​​in each parameter subset of the normal parameter cluster in ascending order to generate a normal ordered sequence, and combine two consecutive electrical parameter values ​​in the sequence into a normal parameter group to be verified. The electrical parameter values ​​in each subset of the abnormal parameter cluster are sorted in descending order to generate an abnormal ordered sequence, and two consecutive electrical parameter values ​​in the sequence are combined into an abnormal parameter group to be verified. Isolated electrical parameter values ​​that do not belong to any parameter subset are not paired and are retained as unassociated parameters.

7. The method for precise calibration based on electricity metering according to claim 6, characterized in that, When performing correlation analysis on each parameter group to be validated using a pre-built parameter correlation model, outputting a correlation score, and counting the number of valid correlation groups with scores exceeding a threshold, the process includes: Collect electrical parameter values, reference parameter values, and deviation data from historical calibration tasks, construct a training dataset, and divide it into a training subset and a test subset; Initialize the graph neural network as a parameter association model, use a training subset to iteratively train the model, and optimize the model parameters through backpropagation; After each iteration, the model's association prediction accuracy is evaluated using a test subset. If the accuracy improvement is less than the minimum increment, the learning rate is adaptively adjusted. When the accuracy converges or reaches the maximum number of iterations, the model parameters are fixed to obtain the trained parameter correlation model. Input the normal and abnormal parameter groups to be verified into the model, output the association probability value of each group, and if the probability value is greater than the association threshold, it is determined to be a valid association group and the number is accumulated.

8. The method for precise calibration based on electricity metering according to claim 7, characterized in that, When determining whether to initiate the calibration process based on a comparison between the overall calibration index and the dynamic calibration threshold, the process includes: Real-time monitoring of environmental factors and workload; dynamic adjustment of calibration thresholds. If the overall calibration index is greater than or equal to the dynamic calibration threshold, a calibration command is generated and the automatic calibration mechanism is activated. If the overall calibration index is less than the dynamic calibration threshold, a maintenance state instruction is generated, and the calibration process is skipped.

9. A precision calibration system based on electricity metering, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the precise calibration method based on electricity metering as described in any one of claims 1 to 8.

Images