Metering error analysis method in electric energy meter drop test

Abstract

The invention discloses a metering error analysis method in an electric energy meter drop test, and the method carries out the electric energy meter drop test through a test site with stable environmental parameters. An electric energy meter to be tested and a standard electric energy meter are fixedly installed on the drop test device, and a remote wireless data communication module is used to connect wiring terminals of the electric energy meters. Enabling the to-be-tested electric energy meter to freely fall through a falling test machine according to preset height and angle, and recording falling height, angle and time parameters; data of the to-be-tested electric energy meter and the standard electric energy meter before falling, during falling and after falling are collected through a sensor; the collected data are preprocessed, and abnormal values and noise data are removed; calculating a metering error in the drop test of the electric energy meter by using a harmonic analysis method; by means of the random forest algorithm and observation of the appearance of the electric energy meter after falling, the error reason is judged, and the metering error analysis capacity in the electric energy meter falling test is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of instruments and power metering, more particularly to a metering error analysis method in a power meter drop test. BACKGROUND

[0002] In the use of the power meter, accidental drop causes changes in the internal structure and circuit parameters of the power meter, loosening, displacement and damage of components; in the traditional measurement method, the impact of drop on the measurement accuracy is not considered, the measurement error caused by drop cannot be accurately analyzed, and the existing technology has problems of not comprehensive consideration of the measurement error and inconsistency of the test conditions with the actual use site in the error analysis of drop. The existing power meter drop test method has many technical limitations: The environmental and working condition parameters are not comprehensive: the traditional test focuses on the basic parameters such as drop height and angle, ignores the influence of environmental and dynamic characteristics such as temperature, humidity and shock waveform on the error, and the test results deviate greatly from the actual scene.

[0003] Data processing and outlier rejection are rough: fixed threshold or single statistical method is used to process collected data, without considering the time sequence characteristics of drop impact (such as the difference between pre-drop stable period and impact transient period), which is easy to misjudge effective data or retain abnormal values, affecting the accuracy of subsequent error analysis.

[0004] Damage modeling lacks virtual-real fusion: relying on physical prototype to detect explicit damage, it is difficult to capture implicit damage such as micro-cracks of solder joints and PCB interlayer peeling, and it is impossible to quantify the dynamic relationship between damage and measurement error, making it difficult to trace the error.

[0005] Error analysis method is single and has poor adaptability: single fundamental wave analysis or linear regression method is mostly used to calculate error, which is difficult to handle complex scenes such as harmonic distortion and multi-parameter coupling, and has insufficient fitting ability for nonlinear error relationship.

[0006] The generalization ability of the evaluation model is weak: the traditional machine learning model relies on real test samples, and the coverage of extreme drop working conditions (such as super-standard height drop) is insufficient, and the model output is not constrained by engineering physics, which is easy to produce prediction results that do not conform to the actual situation.

[0007] To solve the above problems, through the test site with stable environmental parameters, the electric energy meter drop test is carried out; the electric energy meter to be tested and the standard electric energy meter are fixedly installed on the drop test device, and a remote wireless data communication module is used to connect the wiring terminals of the electric energy meter; the electric energy meter to be tested is made to freely fall by the drop test machine according to the preset height and angle, and the height, angle and time parameters of the drop are recorded; the data of the electric energy meter to be tested and the standard electric energy meter before drop, during drop and after drop are collected by sensors; the collected data is preprocessed, abnormal values are removed and noise data is removed; the measurement error in the electric energy meter drop test is calculated by using the harmonic analysis method; and the error cause is judged by the random forest algorithm and observation of the appearance of the electric energy meter after drop. SUMMARY

[0008] In view of the deficiencies in the prior art, the application discloses a measurement error analysis method in an electric energy meter drop test, which can improve the error calculation capability of the electric energy meter in the electric energy meter drop test and improve the error analysis and calculation of the electric energy meter in a special environment.

[0009] A measurement error analysis method in an electric energy meter drop test, comprising the following steps: Step one: measuring the environmental parameters of the test site, recording the environmental parameter data information, and starting the electric energy meter drop test; The environmental parameters at least include temperature of 20-28 DEG C, humidity of 20-50% RH, atmospheric pressure of 101.3 kPa, wind speed, illuminance, drop height, drop times, drop direction and drop mode; Step two: fixing and installing the electric energy meter to be tested and the standard electric energy meter on the electric energy meter calibration device respectively, connecting the wiring terminals of the electric energy meter by using a remote wireless data communication module, measuring the original measurement data information of the electric energy meter, and comparing the error parameters of the electric energy meter by the standard electric energy meter; Then the electric energy meter is dropped from a high altitude, the drop height range is 0-900 cm, the data information of the electric energy meter colliding with the ground when dropping is tested, and the data information at least includes the decibel of the recorded impact sound, the falling distance, angle and time of the electric energy meter; Step three: changing the height or position of the electric energy meter, re-measuring the parameters of the electric energy meter based on the standard table comparison method by the method of step two every time the position is changed, and recording the parameter information of the re-measured electric energy meter; The height points of the electric energy meter to be changed include 300 cm, 700 cm, 110 cm and 450 cm; and the angles include 30 degrees, 60 degrees and 90 degrees; Step four: measuring the dropped electric energy meter by the electric energy meter calibration device again, collecting the electric energy parameter data of the tested dropped electric energy meter by the sensors in the electric energy meter calibration device, and observing the working parameters of the electric energy meter before and after drop; The sensor includes a Hall current sensor and a Hall voltage sensor, and the electric energy meter acquisition parameter includes at least voltage, current, power, power factor and harmonic parameter value; Step five: dynamic damage modeling driven by digital twin and multi-source data fusion; Step six: calculating the electric energy measurement error in the electric energy meter drop test using the harmonic analysis method; Step seven: evaluating the electric energy measurement error after the electric energy meter drop by the random forest algorithm; Wherein the error types in the electric energy measurement error include electric energy parameter error, circuit board damage, display module damage and impact damage; the electric energy parameter error is measured by an error calculation unit, wherein the error calculation unit includes a microprocessor and a data amplifier, an A / D converter, a comparator and a calculator connected with the microprocessor.

[0010] As a further technical solution of the application, in the step two, The remote wireless data communication module is a data communication based on an RS485 communication module, The RS485 communication module includes a differential signal transmission module provided with an A communication end and a B communication end, when the data communication instruction is 1, the voltage on the A communication end is higher than that on the B communication end; when the data communication instruction is 0, the voltage on the A communication end is lower than that on the B communication end, and the receiving end of the differential signal judges the communication instruction by comparing the voltage difference on the A and B communication ends.

[0011] As a further technical solution of the application, in the step three, When the height or position of the electric energy meter is changed, the electric energy meter is dropped in different situations through the first drop mode, the second drop mode and the third drop mode; The first drop mode is free-fall drop; The second drop mode is drop at an angle with the horizontal plane, The third drop mode is artificial weight pressure type damage test after drop; Each time the position is changed, the parameters of the electric energy meter are re-measured based on the standard meter comparison method, and the parameter information of the re-measured electric energy meter is recorded; The height of the electric energy meter is changed to include 300 cm, 700 cm, 110 cm and 450 cm; The angle includes 30 degrees, 60 degrees and 90 degrees.

[0012] As a further technical solution of the application, in the step four, The electric energy meter acquisition parameter processes the measured voltage, current, power, power factor and harmonic parameter values through quartile range; wherein: The step of removing abnormal values by the quartile range is: A: arranging the collected data in order from small to large according to characteristic values, and dividing into three dynamic windows of pre-falling stable section, impact instantaneous section and post-falling recovery section based on the falling test impact timing, and giving different weights to the data of each window; B: calculating the first quartile Q1 and the third quartile Q3 in each dynamic window, wherein Q1 is the 25% quartile value in the window, Q3 is the 75% quartile value in the window, and the window edge data fluctuation influence is eliminated by sliding average correction; C: calculating the dynamic quartile range I QR =Q3-Q1, and introducing an adaptive threshold coefficient k, which dynamically adjusts the value of k with the fluctuation intensity of the window data; D: removing the serious abnormal values less than Q1-k×I QR and greater than Q3+k×I QR in each window, and grading the critical abnormal values, and reserving the corresponding falling impact parameters for subsequent error tracing; E: establishing an abnormal data information association database, and associating the removed serious abnormal values and the marked critical abnormal values with the impact intensity, duration and other parameters of the falling test. As a further technical solution of the present application, in step five, the method of dynamic damage modeling driven by digital twin and multi-source data fusion is: Step 51: multi-dimensional data preprocessing and twin initialization Collecting the high-precision multi-dimensional physical impact data in step two and the dynamic damage information in step four, and realizing the synchronization association of physical data and visual data through a space-time alignment algorithm; taking the preprocessed data as input, constructing a digital twin containing the internal structure of the electric energy meter, material properties and boundary conditions, initializing the mechanical parameters and geometric model of the virtual scene; Step 52: explicit dynamics simulation and implicit damage visualization prediction Based on the explicit dynamics algorithm, the digital twin is driven to reproduce the falling impact process, dynamically outputting the stress / strain distribution cloud of the key structure, displacement trajectory and material yield threshold overrun area, and calculating the damage degree through a damage accumulation factor, wherein the damage accumulation factor calculation function is: In formula (1), is the instantaneous stress, is the material yield strength, is the stress duration; step 53: error analysis by error association model of multi-source data fusion A "damage-error" mapping database is constructed to associate the high-risk damage area with the corresponding parameter error calculation; a multiple linear regression model is optimized, and a damage factor is introduced as an independent variable, and the model formula is: In formula (2), H is the drop height, A is the angle, N is the number, T is the temperature, M is the humidity, DCF is the damage cumulative factor, is the regression coefficient, is the error term; step 54: dynamic correction of test error The measured value of the metrological error of the physical prototype is fed back to the digital twin, the simulation parameters are adjusted through the self-adaptive correction algorithm, the matching degree of the damage area predicted by the twin and the actual test result is improved to more than 90%, and virtual iterative tests are carried out based on the corrected model to simulate the damage evolution and error change trend under different drop conditions; Step 55: damage traceability and error prediction in the whole life cycle A damage traceability knowledge base is constructed, and the simulation data of the twin, the multi-source fusion model parameters and the error measured value of each test are stored in the blockchain to form an unalterable test file. Through knowledge graph technology, historical data is associated, when new drop parameters are input, the damage-error model of similar working conditions is quickly called to realize the advance prediction of metrological error and locate the key influencing factors of potential damage. As a further technical solution of the present application, the calculation steps of the harmonic analysis method in step six are: step 61: dynamic time domain signal acquisition and preprocessing The voltage (u(t)) and current (i(t)) signals containing the fundamental and harmonic components in the drop test of the electric energy meter are collected, and the sampling frequency is set to 256 times the fundamental frequency to meet the Nyquist criterion; based on the impact instantaneous period output by the digital twin in step 52, the time domain signal is segmented and weighted for preprocessing; Step 62: improved Fourier transform and harmonic parameter separation The preprocessed voltage and current signals are subjected to windowed short-time Fourier transform to realize time-frequency domain conversion, and the Hanning window is selected to suppress spectral leakage, and the window length is dynamically adjusted according to the harmonic number; the transformation formula is: In formula (3), is the Hanning window function, is the correction coefficient of drop height on spectral amplitude, is the correction coefficient of temperature on phase shift, h is the drop height, and T is the environmental temperature; step 63: harmonic power calculation with multiple parameter coupling A dynamic correction model of harmonic power is constructed based on the damage cumulative factor output by the digital twin, and the calculation formula is: In formula (4), a new damage correction term quantifies the influence of implicit damage on harmonic transmission; a and b are the influence degrees of drop height (h) and temperature (T) on the harmonic power coefficient, respectively The harmonic layering weighting method is used to calculate the electric energy measurement error, and the weights are allocated according to the contribution of each harmonic to the total error, and the error calculation formula is: In formula (5), k is the harmonic weight, is the measurement period, is the theoretical electric energy value calculated based on the harmonic power in the period, are the electric energy meter readings at t1 and t2, respectively; step 65: dynamic calibration of error results The calculated harmonic measurement error and the error prediction value output by the multiple linear regression model in step 53 are fused and calibrated, and the Kalman filtering algorithm is used to iteratively optimize the error results: the harmonic analysis method result is used as the observation value, the regression model result is used as the prediction value, the weights of the two are dynamically adjusted, and the calibrated measurement error value is output. As a further technical solution of the present application, The working method of the random forest algorithm is: step 71: physical correlation feature extraction and fusion, extracting multi-dimensional core features and constructing correlation mapping Extract the electric energy meter internal circuit structure parameters, component electrical characteristics, mechanical part size, and different time measurement value change characteristics; bind the circuit structure parameters with the damage area output by the digital twin; Step 72: hierarchical sampling and physical perception feature selection Damage hierarchical sampling is used to stratify the original data according to the damage degree in the drop test, and Bootstrap sampling with replacement is performed within each layer to ensure that the sample proportion of each damage level is consistent with the actual working condition; When randomly selecting features, introduce a physical correlation screening mechanism to preferentially select splitting features from the feature subset that is strongly correlated with the measurement error; Step 73: dynamic constraint decision tree growth and integration The decision tree growth adopts a double-rule splitting mechanism: the minimum Gini coefficient is used as the basic criterion, and a physical constraint condition is introduced, when the splitting node involves a key parameter, the parameter is forced to be used as the splitting basis, to ensure that the decision tree growth conforms to the engineering physical law; 100 independent decision trees are generated according to the predetermined requirements, each tree is grown based on different sampling samples and feature subsets, and finally the random forest model is integrated; Step 74: multi-type weighted voting prediction and error type determination Input new sample data to each decision tree for parallel prediction: mean integration is used for metering error value, and weighted voting mechanism is used for error type, and weights are assigned according to the prediction accuracy of each decision tree in historical data; based on the error type counting formula: In formula (6), T is the prediction error type, is an indicator function, is the output of the jth tree for the ith error type, is the actual prediction result of the jth tree, the error type with the highest votes is screened, and the confidence of the type is output; step 75: cross-condition verification and model dynamic optimization The power meter data under different test conditions are used for verification, the error prediction deviation under each condition is calculated, the model is optimized when the deviation is greater than 5%, and the decision tree splitting rule is adjusted by increasing the sample under the corresponding condition; based on the verification result, an adaptive table is constructed to realize accurate identification of the difference of metering error influence under cross-condition.

[0013] Active and beneficial technical effects: The present application can realize accurate reproduction and high-quality data acquisition of real use scenarios by designing a full-scene simulation scheme of multi-environment parameter recording, multi-mode drop test and gradient working condition setting, combining the high anti-interference characteristics of RS485 differential signal transmission, and providing a reliable basis for subsequent analysis; Through time sequence segmentation weighting of the dynamic window quartile range algorithm, adaptive threshold abnormal value grading and abnormal value-impact associated database construction, dynamic processing and accurate identification of abnormal values can be realized, and error tracing of the power meter can be performed; Through explicit dynamic simulation of the digital twin, damage cumulative factor (DCF) calculation and "damage-error" mapping model optimization, visual prediction and damage-error correlation quantification of implicit damage can be realized; Through harmonic separation of windowed short-time Fourier transform (STFT), power model of damage correction term fusion, hierarchical weighted error calculation and Kalman filter calibration, the influence of drop damage on harmonic distortion can be accurately quantified; Through physical correlation feature fusion, damage hierarchical sampling and random forest model construction of double-rule decision tree growth, combined with virtual sample enhancement and cross-condition verification, intelligent evaluation of error under extreme working conditions can be realized. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiment or prior art description will be briefly introduced below, and obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Figure 1 A flowchart of a metering error analysis method in a power meter drop test of the present application is shown in the figure; Figure 2 A flowchart of a step of removing outliers from the interquartile range is shown in the figure; Figure 3 A flowchart of a method of digital twin driven dynamic damage modeling and multi-source data fusion of the present application is shown in the figure; Figure 4 A flowchart of a calculation step of the harmonic analysis method of the present application is shown in the figure; Figure 5 A flowchart of the working method of the random forest algorithm of the present application is shown in the figure. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments will be described in detail below with reference to the accompanying drawings in the embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, but not all. It should be understood that the description is only exemplary, but not to limit the scope of the present application. In addition, in the description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present application.

[0016] A metering error analysis method in a power meter drop test, comprising the following steps: Step 1: Measure the environmental parameters of the test site, record the environmental parameter data information, and start the power meter drop test; The environmental parameters include at least temperature of 20-28℃, humidity of 20-50%RH, atmospheric pressure of 101.3 kPa, wind speed, illuminance, drop height, drop frequency, drop direction and drop method; Step 2: Fix and install the power meter to be tested and the standard power meter on the power meter calibration device respectively, and connect the wiring terminals of the power meter using a remote wireless data communication module, measure the original metering data information of the power meter, and compare the error parameters of the power meter with the standard power meter; Then drop the power meter from a high altitude, the drop height range is 0-900 cm, test the data information of the power meter when it collides with the ground, the data information at least includes the decibel of the impact sound, the falling distance, the angle and the time of the power meter; Step 3: Change the height or position of the power meter, and each time the position is changed, the method of step 2 is used to measure the parameters of the power meter based on the standard meter comparison method, and record the parameter information of the power meter measured again; The height points of the power meter to be changed include 300 cm, 700 cm, 110 cm and 450 cm; the angles include 30 degrees, 60 degrees and 90 degrees; Step four: measure the dropped electric energy meter again by the electric energy meter calibration device, collect the electric energy parameter data of the electric energy meter to be tested after dropping through the sensor in the electric energy meter calibration device to observe the working parameters of the electric energy meter before and after dropping; The sensor includes a Hall current sensor and a Hall voltage sensor, and the electric energy meter acquisition parameters at least include voltage, current, power, power factor and harmonic parameter values; Step five: dynamic damage modeling and multi-source data fusion driven by digital twin; Step six: calculate the electric energy measurement error in the electric energy meter drop test using the harmonic analysis method; Step seven: evaluate the electric energy measurement error after the electric energy meter drops through the random forest algorithm; Wherein the error types in the electric energy measurement error include electric energy parameter error, circuit board damage, display module damage and impact damage; the electric energy parameter error is measured by an error calculation unit, wherein the error calculation unit includes a microprocessor and a data amplifier, an A / D converter, a comparator and a calculator connected with the microprocessor.

[0017] In the step two, the remote wireless data communication module is a data communication based on an RS485 communication module, The RS485 communication module contains a differential signal transmission module provided with an A communication end and a B communication end, when the data communication instruction is 1, the voltage on the A communication end is higher than that on the B communication end; when the data communication instruction is 0, the voltage on the A communication end is lower than that on the B communication end, the receiving end of the differential signal judges the communication instruction by comparing the voltage difference on the A and B communication ends.

[0018] In the above embodiment, an industrial-grade RS485 communication module is selected and integrated in the communication interface unit of the electric energy meter calibration device. The module is provided with an independent A communication end (non-reverse signal) and a B communication end (reverse signal), which are respectively connected to the RS485 wiring terminals of the electric energy meter to be tested and the standard electric energy meter through shielded twisted pair wires, and the shielded layer is single-ended grounded to reduce electromagnetic interference.

[0019] In module A, B communication end series TVS transient suppression diode and self-recovery fuse, prevent the damage of impact voltage or short circuit to module in drop test; Through optical coupling isolation circuit realizes electrical isolation of module and electric energy meter mainboard, sets communication parameter through module configuration software - baud rate 9600bps (adapt to electric energy meter measurement data transmission rate), data format is '8 data bits + 1 stop bit + no check bit', and working mode is half duplex (supports master-slave communication). Define the address of master station (test device) as 0x01, the address of electric energy meter to be tested as 0x02, and the address of standard electric energy meter as 0x03, realize parallel communication of multiple devices through address distinction. In module firmware, preset signal logic rule - when transmission data communication instruction is '1' (high level), A communication end output voltage (3-5V) is higher than B communication end voltage (0-0.5V), and voltage difference is greater than or equal to 2V; When the instruction is '0' (low level), A communication end voltage (0-0.5V) is lower than B communication end voltage (3-5V), and voltage difference is less than or equal to -2V. The built-in comparator in the receiving end realizes instruction decoding by judging the sign and absolute value of the voltage difference between A and B.

[0020] After the test starts, the test device sends acquisition instructions to the electric energy meter to be tested and the standard electric energy meter through the RS485 module in a cycle, and the instruction frame format is'start code + device address + instruction type + data length + check code + end code'. For example, the instruction frame for collecting voltage parameters is '0xAA 0x02 0x01 0x01 0x55 0xBB' (0x02 is the address of the electric energy meter to be tested, and 0x01 is the voltage acquisition instruction). After the electric energy meter receives the instruction, it encodes the measurement data such as voltage and current into binary signals, and converts them into differential signals through the internal RS485 interface: data '1' corresponds to A>B voltage difference, and data '0' corresponds to A<B voltage difference. The signal is transmitted to the RS485 module of the test device through a twisted pair, and the receiving end comparator samples the voltage difference between A and B in real time, restores the differential signal to binary data, and then analyzes the actual measurement value after verification (CRC cyclic redundancy check).

[0021] When the electric energy meter performs drop test, the module automatically enables anti-interference mechanism, increases signal driving current to improve voltage difference stability, adopts Manchester encoding to eliminate DC component interference, and the test device real-time statistics RS485 communication frame error rate, retransmission times and other indicators, when receiving fails for 3 times in succession, automatically switches to standby baud rate (4800bps) and increases check bit (odd check), tries to recover communication link. The built-in state monitoring circuit in the module triggers a hardware alarm signal when A / B end short circuit, voltage difference is abnormal (|ΔV|<1V) or communication timeout (>100ms), the test device records fault time, fault type and displays 'RS485 communication exception', and starts local storage mechanism to store measurement data, and transmits data after communication recovery.

[0022] In the third step, the electric energy meter is dropped in different situations through the first drop mode, the second drop mode and the third drop mode when the height or position of the electric energy meter is changed; The first drop mode is free-fall drop; The second drop mode is drop at an angle with the horizontal plane, The third drop mode is artificial weight crushing and damage test after drop; Each time the position is changed, the parameters of the electric energy meter are measured again based on the standard meter comparison method through the method of step two, and the parameter information of the electric energy meter measured again is recorded; The height points of the electric energy meter include 300 cm, 700 cm, 110 cm and 450 cm; The angles include 30 degrees, 60 degrees and 90 degrees.

[0023] In the above embodiment, through the combination design of multi-mode drop and gradient working conditions, this method realizes full-scene coverage from normal scene to extreme scene, from slight damage to serious damage, and combines standard meter comparison and data reproduction mechanism, which not only can accurately capture error characteristics corresponding to different damage types, but also can quantify the correlation between impact energy, angle and error, providing full- range test basis for structure optimization, error tracing and grade division of electric energy meter, and significantly improving the scientificity and engineering practical value of measurement error analysis.

[0024] In the fourth step, the electric energy meter collects parameters and processes the measured voltage, current, power, power factor and harmonic parameter values through quartile range; wherein: The step of removing outliers by quartile range is: A: arrange the collected data in the order of characteristic value from small to large, and divide into three dynamic windows of pre-drop stable section, impact instantaneous section and post-drop recovery section based on drop test impact timing, and give different weights to the data of each window; B: calculate the first quartile Q1 and the third quartile Q3 in each dynamic window, wherein Q1 is the 25% quartile value in the window, Q3 is the 75% quartile value in the window, and the sliding average correction is used to eliminate the influence of window edge data fluctuation; C: calculate the dynamic quartile range I QR =Q3-Q1, and introduce an adaptive threshold coefficient k, which dynamically adjusts with the fluctuation intensity of window data; D: remove the data smaller than Q1-k×I QR and greater than Q3+k×I QRsevere outliers while classifying critical outliers, and reserving their corresponding drop impact parameters for subsequent error tracing; E: Establishing an abnormal data information correlation database, correlating the removed severe outliers and the labeled critical outliers with the impact strength, duration, and other parameters of the drop test.

[0025] In the above embodiment, based on the impact physical characteristics of the drop test, the time-domain data of the collected voltage, current, power, and other parameters are divided into three dynamic windows: Pre-drop stable segment: stable running data 5-10 seconds before the drop occurs, reflecting the parameter baseline value under the normal working state of the electric energy meter; Impact instantaneous segment: dramatic fluctuation data at the impact moment and the subsequent 0.5-2 seconds, containing the parameter mutation characteristics caused by the impact; Post-drop recovery segment: parameter recovery period data 2-10 seconds after the impact ends, reflecting the process of the electric energy meter transitioning from an abnormal state to a stable state.

[0026] According to the importance of each window data to error analysis, weights are assigned to the impact instantaneous segment, the pre-drop stable segment, and the post-drop recovery segment, with the weights being 0.5, 0.3, and 0.2, respectively. The influence of the key period data of the impact is strengthened through weighted calculation.

[0027] Quartile calculation and edge correction (step B): 1. Accurate extraction of quartiles: In each dynamic window, the data is sorted from small to large according to the characteristic value, and the first quartile (Q1) and the third quartile (Q3) are calculated by linear interpolation method: Q1 is the characteristic value at the 25% position in the window, representing the lower quartile level in the data set; Q3 is the characteristic value at the 75% position in the window, representing the upper quartile level in the data set.

[0028] 2. Sliding average edge correction: To eliminate the interference of instantaneous fluctuations on quartile calculation, a 3-5 point sliding average algorithm is used to correct Q1 and Q3, ensuring that Q1 and Q3 can truly reflect the distribution characteristics of the data in the window. 1. Interquartile range (IQR) calculation: The data dispersion is calculated by the formula IQR=Q3-Q1 in each window. The larger the IQR value, the more dramatic the parameter fluctuations in the window (such as the IQR of the impact instantaneous segment is usually 3-5 times that of the stable segment).

[0029] 2. Dynamic adjustment of threshold coefficient k: Introduce the window data fluctuation standard deviation (σ) as the basis for judgment. When σ ≥ 0.05 times the window data mean (intense fluctuation, such as impact transient period), set k to 1.8 to strictly remove outliers; when σ < 0.05 times the mean (gentle fluctuation, such as stable period), set k to 1.2 to retain more potential valid data, achieving adaptive matching of threshold and data characteristics.

[0030] Abnormal value grading processing and parameter association (step D) 1. Severe outlier removal: Mark and remove extreme outliers less than Q1-k×IQR and greater than Q3+k×IQR in each window. Such data is often caused by sensor transient failure, loose wiring, and other serious disturbances caused by impact (such as voltage drop to zero, current spike pulse). Removing them can avoid misleading subsequent error calculations.

[0031] 2. Critical outlier labeling: Grade critical outliers in the range of Q1-1.2k×IQR to Q1-k×IQR, Q3+k×IQR to Q3+1.2k×IQR (mild anomaly / moderate anomaly), and record the corresponding drop impact parameters (such as critical outlier associated impact acceleration peak, angle in impact transient period).

[0032] Abnormal database construction and traceability support (step E) 1. Database structured storage: Establish an abnormal data information association database containing fields such as "window type, abnormal value level, parameter name, abnormal value size, corresponding impact parameters (height / angle / acceleration), timestamp", etc., to achieve standardized management of abnormal data.

[0033] 2. Multi-dimensional correlation analysis: Through database queries, abnormal parameters under specific drop conditions can be quickly located (such as moderate anomaly in current parameters during impact transient period under 700 cm height, 60° angle drop), combined with subsequent digital twin damage modeling (step five) and harmonic analysis (step six), the physical root cause of abnormal values (such as current sampling distortion caused by PCB board solder crack) can be traced, providing data support for accurate traceability of measurement errors. In step five, the method of dynamic damage modeling driven by digital twin and multi-source data fusion is as follows: Collect high-precision multi-dimensional physical impact data in step two and dynamic damage information in step four, and synchronize the association of physical data and visual data through space-time alignment algorithm; input the preprocessed data to build a digital twin containing the internal structure, material properties and boundary conditions of the electric energy meter, and initialize the mechanical parameters and geometric model of the virtual scene; Step 52: Explicit dynamics simulation and implicit damage visualization prediction The drop impact process is reproduced based on an explicit dynamic algorithm driven digital twin, and the stress / strain distribution cloud, displacement trajectory and material yield threshold overrun area of the key structure are dynamically output, the injury degree is calculated through a damage accumulation factor, wherein the damage accumulation factor calculation function is: In formula (1), is the instantaneous stress, is the material yield strength, is the stress duration; step 53: error analysis by error correlation model of multi-source data fusion A "damage-error" mapping database is constructed, the high-risk damage area is associated with the corresponding parameter for error correlation calculation; a multiple linear regression model is optimized, a damage factor is introduced as an independent variable, and the model formula is: In formula (2), H is the drop height, A is the angle, N is the number, T is the temperature, M is the humidity, DCF is the damage accumulation factor, is the regression coefficient, is the error term; step 54: dynamic correction of test error The measurement error of the physical prototype is fed back to the digital twin, the simulation parameters are adjusted by the self-adaptive correction algorithm, the matching degree of the damage area predicted by the twin and the actual test result is improved to more than 90%, and the virtual iterative test is carried out based on the corrected model to simulate the damage evolution and error change trend under different drop conditions; Step 55: damage traceability and error prediction in the whole life cycle A damage traceability knowledge base is constructed, the twin simulation data of each test, the multi-source fusion model parameters and the error measured value are stored in the blockchain to form an unalterable test file, and the historical data are associated through knowledge graph technology, when new drop parameters are input, the damage-error model of similar conditions is quickly called to realize the advance prediction of the measurement error and locate the key influencing factors of potential damage. The specific effects of the present application are verified through specific tests, three groups of different specifications of electric energy meter prototypes (single-phase intelligent meter, three-phase intelligent meter and Internet of Things electric energy meter) are selected, and tests are carried out in the multi-condition drop test environment constructed in steps one to four, covering the following key conditions: The physical impact data (acceleration, impact time) of step two, the dynamic damage information (shell deformation, parameter abnormal value) of step four and the measured value of the measurement error are synchronously collected as the input and verification benchmark of the digital twin modeling.

[0034] Through the explicit dynamic simulation of step 52, the stress / strain cloud and the damage accumulation factor (DCF) output by the digital twin can accurately locate the hidden damage: Test Group 1 (700cm free fall): Simulation predicted a DCF of 0.92 (≥0.8 indicates high risk) in the PCB metrology chip solder joint area, corresponding to a strain of 1800με (exceeding the material yield threshold of 1500με). Actual disassembly and inspection revealed two microcracks in this area, with a 100% prediction accuracy.

[0035] Compared with traditional methods: Visual inspection alone fails to detect hidden damage and requires destructive disassembly. Digital twins can predict hidden damage non-invasively, improving detection efficiency by 80%.

[0036] Step 53 introduces the damage accumulation factor (DCF) to optimize the multivariate linear regression model and compares the error prediction effect of the traditional model without DCF: In test group 2 (450cm angle drop), the traditional model failed to account for PCB lateral bending damage (DCF = 0.75), resulting in a prediction error of +1.2%, significantly deviating from the measured error of +0.5%. After DCF correction, the proposed method reduced the prediction error to +0.6%, reducing the deviation to within 0.1%.

[0037] The adaptive correction algorithm in step 54 significantly improved the match between the twin predictions and actual damage. In the initial simulation, the predicted shell deformation for test group 3 deviated by 12% from the measured value. After incorporating feedback correction based on the measured metrological error of the physical prototype (+2.1%) and adjusting the material elastic modulus parameter (from 200 GPa to 185 GPa), the secondary simulation error dropped to 3.5%, improving the match to 96.5%. Based on this corrected model, virtual iterative tests were conducted, simulating 100 operating conditions not tested physically (e.g., 500 cm, 45° drop). The predicted error was ≤0.5%, reducing the use of physical prototypes by 60%.

[0038] The blockchain evidence storage and knowledge graph traceability mechanism in step 55 has a 100% hash verification consistency rate for the test data stored in the blockchain, eliminating the risk of data tampering. When entering new working condition parameters (600cm, 75° drop), the knowledge graph matches a similar historical working condition (test group 1) within 1.2 seconds, and calls its damage-error model. The prediction error is +0.8%, which is only 0.05% different from the subsequent physical test result of +0.75%. The traceability response speed is 90% faster than traditional database queries. The calculation steps of the harmonic analysis method described in step 6 are: Step 61: Dynamic time domain signal acquisition and preprocessing The voltage (u(t)) and current (i(t)) signals containing fundamental and harmonic components during the electric energy meter drop test are collected, with the sampling frequency set to 256 times the fundamental frequency to meet the Nyquist criterion. Based on the impact transient period output by the digital twin in step 52, the time domain signals are preprocessed by segmented weighting. Step 62: Improved Fourier transform and harmonic parameter separation The pre-processed voltage and current signals are converted from time domain to frequency domain by using windowed short-time Fourier transform, and the Hanning window is selected to suppress spectral leakage, and the window length is dynamically adjusted according to the harmonic order; the transformation formula is: In formula (3), is the Hanning window function, is the correction coefficient of the drop height on the spectral amplitude, is the correction coefficient of the temperature on the phase shift, h is the drop height, and T is the environmental temperature; Step 63: Harmonic power calculation with multiple parameter coupling A dynamic correction model of harmonic power is constructed based on the damage cumulative factor output by the digital twin, and the calculation formula is: In formula (4), a new damage correction term is added to quantify the influence of hidden damage on harmonic transmission; a and b are the influence degrees of the drop height (h) and the temperature (T) on the harmonic power coefficient, respectively; Step 64: Hierarchical measurement error calculation and traceability The harmonic hierarchical weighting method is used to calculate the electric energy measurement error, and the weights are allocated according to the contribution of each harmonic to the total error, and the error calculation formula is: In formula (5), is the weight of the kth harmonic, is the measurement period, is the theoretical electric energy value calculated based on the harmonic power in this period, are the electric energy meter readings at t1 and t2, respectively; Step 65: Dynamic calibration of error results The calculated harmonic measurement error and the error prediction value output by the multiple linear regression model in step 53 are fused and calibrated, and the Kalman filter algorithm is used to iteratively optimize the error results: the harmonic analysis method result is used as the observation value, the regression model result is used as the prediction value, the weights of the two are dynamically adjusted, and the calibrated measurement error value is output. In the above embodiment, select 3 The same model of smart energy meters (numbered A, B, and C) were tested under different drop conditions: Drop conditions: height 300 cm / 700 cm, angle 30° / 60°, covering the first drop mode (free fall) and the second drop mode (angle drop); Data acquisition: collect the voltage (u(t)) and current (i(t)) signals after dropping through Hall voltage / current sensors, with a sampling frequency of 12.8 kHz (256 times of the fundamental frequency 50 Hz), and simultaneously record the environmental temperature (23±2℃), humidity (35±5%RH), and the damage cumulative factor (DCF) output by the digital twin.

[0039] ·Original signal contrast: The impact transient period signal without preprocessing contains a large amount of high-frequency noise (signal-to-noise ratio about 25 dB), after segmented weighted preprocessing (impact transient period weight 0.5), the signal-to-noise ratio is improved to 42 dB, and the waveform smoothness is significantly improved; ·Fourier transform accuracy: Using the improved STFT with Hanning window (window length adjusted according to harmonic order: fundamental wave 20 ms, 5th harmonic 5 ms), the amplitude separation error of 3rd, 5th and 7th harmonics is less than 2%, and the phase difference measurement error is less than 1°, which is significantly lower than the amplitude error (5-8%) of traditional fixed window length FFT. Based on the DCF output of digital twin (0.3-0.8, corresponding to slight to moderate damage), a dynamic correction model is constructed, and the traditional power calculation results without introducing damage correction term are compared: Analysis: The damage correction term (1+0.5·DCF) effectively quantifies the influence of micro-cracks in solder joints and PCB deformation on harmonic transmission, reducing the power calculation error by an average of 65%, especially in high damage conditions (DCF>0.7).

[0040] Using hierarchical weighting method (3rd harmonic weight 0.3, 5th harmonic 0.2, 7th and above 0.5) to calculate error and associate damage area: Analysis: The hierarchical weighting method focuses on high-contribution harmonics, improving the error calculation accuracy by more than 55%; combined with the "harmonic-damage" correlation, it can directly locate the error source (such as 3rd harmonic anomaly corresponding to voltage sampling loop damage), and the tracing efficiency is improved by 40% compared to traditional full harmonic average method.

[0041] Fuse the results of harmonic analysis method and the results of multivariate linear regression model in step 53 (Kalman filter weight dynamic adjustment): Analysis: After calibration, the error deviation is significantly reduced, and the 95% confidence interval is narrowed from ±2.0% to ±1.2%, and the result stability meets the high-precision requirements of electric energy metering error detection (error allowed range ±2%).

[0042] The improved STFT and dynamic window length design solve the spectrum leakage problem of traditional FFT, and the harmonic parameter measurement error is less than 2%; the DCF damage correction term is introduced, the influence of hidden damage on harmonic power is accurately captured, and the power calculation error is reduced by 65%; the hierarchical weighting method and the "harmonic-damage" correlation model realize the directional positioning of the error root, and the tracing efficiency is improved by 40%; the multi-method fusion calibration reduces the error deviation by 41%, which provides core technical support for accurate analysis of measurement error in drop test, and the comprehensive performance is improved by more than 50% compared with the traditional single fundamental analysis method. It is proved that the scheme has outstanding effects. In further embodiments, in step seven, the working method of the random forest algorithm is: step 71: physical correlation feature extraction and fusion, extracting multi-dimensional core features and constructing correlation mapping Extracting the internal circuit structure parameters of the electric energy meter, the electrical characteristics of the components, the size of the mechanical parts, and the change characteristics of the measurement values at different times; binding the circuit structure parameters with the damage area output by the digital twin; Step 72: hierarchical sampling and physical perception feature selection Damage hierarchical sampling is used to divide the original data according to the damage degree in the drop test, and Bootstrap sampling with replacement is performed in each layer to ensure that the sample proportion of each damage level is consistent with the actual working condition; when randomly selecting features, a physical correlation screening mechanism is introduced to preferentially select splitting features from the feature subset with strong correlation with measurement error; Step 73: dynamic constraint decision tree growth and integration The decision tree growth adopts a double-rule splitting mechanism: taking the minimum Gini coefficient as the basic criterion, and introducing a physical constraint condition, when the splitting node involves a key parameter, the parameter is forced to be used as the splitting basis to ensure that the decision tree growth conforms to the engineering physical law; 100 independent decision trees are generated according to the predetermined requirements, each tree is grown based on different sampling samples and feature subsets, and finally the random forest model is formed by integration; Step 74: multi-type weighted voting prediction and error type determination Input new sample data into each decision tree for parallel prediction: use mean integration for measurement error value, and use weighted voting mechanism for error type, according to the prediction accuracy of each decision tree in historical data; based on the error type counting formula: In formula (6), T is the predicted error type, is an indicator function, is the output of the jth tree for the ith error, is the actual prediction result of the jth tree, the error type with the highest votes is selected, and the confidence of the type is output; step 75: cross-condition verification and model dynamic optimization The data of the electric energy meter under different test conditions is verified, the error prediction deviation under each condition is calculated, and when the deviation is greater than 5%, the model optimization is triggered, and the decision tree splitting rule is adjusted by increasing the sample under the corresponding condition for retraining; based on the verification result, an adaptation table is constructed to realize the accurate identification of the difference of the measurement error influence under different conditions. I will present the implementation effect of the random forest algorithm in the measurement error analysis of the electric energy meter drop test through specific test data and comparative analysis, to verify its technical advantages.

[0043] Select 50 intelligent electric energy meters of the same type to carry out drop tests under multiple working conditions: Drop parameters: height 110 cm / 300 cm / 450 cm / 700 cm, angle 30° / 60° / 90°, covering free fall, angle drop, and heavy object pressure; Feature dataset: 86-dimensional features are extracted from steps four / five / six, including circuit structure parameters (PCB board thickness, solder joint spacing), component characteristics (measurement chip precision level), dynamic damage features (DCF value, stress concentration coefficient), harmonic parameters (3 / 5 / 7 harmonic distortion rate), and environmental parameters (temperature, humidity), which are divided into training set (35) and test set (15) according to 7:3.

[0044] In the implementation process: 1. Feature extraction and fusion (step 71): Through physical correlation screening, 42-dimensional core features are retained, and an "circuit parameter-damage area" correlation mapping is constructed (such as binding "solder joint spacing" to the "voltage sampling circuit damage area" of digital twin positioning), the average correlation is improved by 35%.

[0045] 2. Sampling and feature selection (step 72): According to the damage degree (mild / medium / severe, DCF=0.2-0.4 / 0.4-0.7 / 0.7-1.0), stratified sampling is performed, and each layer sample accounts for 40% / 35% / 25%; When randomly selecting features, prefer to select from the "DCF value-harmonic distortion rate-power error" strong correlation subset (correlation>0.7), the feature selection efficiency is improved by 28%.

[0046] 3. Model construction (step 73): Generate 100 decision trees, set the maximum depth to 15 layers, and the minimum leaf node sample size to 5; The double rule splitting mechanism is as follows: based on the minimum Gini coefficient, when involving key parameters such as "measurement chip pin damage" and "3rd harmonic phase difference", forced priority splitting is performed to ensure that the decision logic conforms to the "damage-error" physical transmission law.

[0047] 4. Prediction and optimization (steps 74-75): The test set 15 power meters are predicted, and the error type is divided into four categories: power parameter error, circuit board damage, display module damage, and impact damage. When the prediction deviation is greater than 5%, the corresponding working condition sample (such as 700 cm / 60° heavy weight pressure) is supplemented for retraining, and the model is iteratively optimized for 3 rounds. Analysis: Random forest solves the problem of insufficient fitting of traditional methods to nonlinear error relationships through multi-tree integration and physical constraints. In particular, in moderate / serious damage conditions, the prediction accuracy advantage is more significant (accuracy rate of 90.5% when error > 1%, linear regression only 62.1%). Analysis: Through feature importance, the main cause of error can be directly located (such as a sample error of 92% caused by "DCF=0.85+3 harmonic distortion rate=5.2%"), and the tracing efficiency is improved by 60% compared to the traditional full feature traversal method, and the matching degree with the digital twin damage positioning result is 91%.

[0048] Performance under extreme unseen conditions (such as 900 cm super-standard height drop): Analysis: After introducing digital twin virtual samples for enhancement, the model's adaptability to extreme conditions is significantly improved, with a deviation rate of less than 5%, which is more stable than the unenhanced model (deviation rate of 12-25%). Analysis: Weighted voting based on decision tree accuracy (such as giving a tree with a historical accuracy of > 90% a weight of 1.5 times) effectively reduces the misjudgment rate, especially for low-frequency error types such as "display module damage", the determination accuracy is improved most significantly.

[0049] Random forest algorithm realizes high-precision analysis and efficient tracing of power meter drop test measurement error through physical correlation feature fusion, dynamic constraint modeling, and weighted voting mechanism, and the technical superiority is significant.

[0050] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these specific embodiments are only illustrative, and those skilled in the art can make various omissions, substitutions and changes to the details of the above method and system without departing from the principles and essence of the present application. For example, combining the above method steps, performing substantially the same function to achieve substantially the same result according to the substantially same method is within the scope of the present application. Therefore, the scope of the present application is only limited by the appended claims.

Claims

1. A method for analyzing measurement errors in an electric energy meter drop test, characterized by: The following steps are involved: Step 1: Measure the environmental parameters of the test site, record the environmental parameter data information, and start the electric energy meter drop test; The environmental parameters include at least a temperature of 20°C to 28°C, a humidity of 20-50%RH, an atmospheric pressure of 101.3 kPa, wind speed, illumination, drop height, number of drops, drop direction, and drop method; Step 2: The energy meter to be tested and the standard energy meter are fixedly installed on the energy meter calibration device respectively, and the long-range wireless data communication module is used to connect the wiring terminals of the energy meter to measure the original metering data information of the energy meter, and the error parameters of the energy meter are compared with the standard energy meter; Then, the electric energy meter is dropped from a height of 0-900 cm to test data information of the electric energy meter colliding with the ground when it falls, and the data information at least includes the decibel of the impact sound, the falling distance, angle and time of the electric energy meter; Step 3: Change the height or position of the electric energy meter. Each time the position is changed, re-measure the parameters of the electric energy meter based on the standard meter comparison method according to the method in step 2, and record the parameter information of the re-measured electric energy meter; The height points for changing the electric energy meter include 300 cm, 700 cm, 110 cm and 450 cm; the angles include 30 degrees, 60 degrees and 90 degrees; Step 4: Measure the dropped electric energy meter again using the electric energy meter calibration device, and collect the electric energy parameter data after the test drop through the sensor in the electric energy meter calibration device to observe the working parameters of the electric energy meter before and after the drop; The sensors include a Hall current sensor and a Hall voltage sensor, and the electric energy meter collects parameters including at least voltage, current, power, power factor and harmonic parameter values; Step 5: Digital twin-driven dynamic damage modeling and multi-source data fusion; Step 6: Use harmonic analysis to calculate the energy measurement error during the energy meter drop test; Step 7: Use the random forest algorithm to evaluate the energy metering error after the energy meter drops; The error types in the electric energy metering error include electric energy parameter error, circuit board damage, display module damage and impact damage; the electric energy parameter error is measured by an error calculation unit, wherein the error calculation unit includes a microprocessor and a data amplifier, an A / D converter, a comparator and a calculator connected to the microprocessor.

2. A method for analyzing measurement errors in an electric energy meter drop test according to claim 1, characterized in that: In the step 2, The remote wireless data communication module is based on the data communication of RS485 communication module. The RS485 communication module includes a differential signal transmission module provided with an A communication terminal and a B communication terminal. During data communication, when the data communication instruction is 1, the voltage on the A communication terminal is higher than the voltage on the B communication terminal; when the data communication instruction is 0, the voltage on the A communication terminal is lower than the voltage on the B communication terminal. The receiving end of the differential signal determines the communication instruction by comparing the voltage difference between the A and B communication terminals.

3. A method for analyzing measurement errors in an electric energy meter drop test according to claim 1, characterized in that: In the step three, When the height or position of the electric energy meter is changed, the electric energy meter is dropped in different scenarios through the first drop mode, the second drop mode, and the third drop mode; The first drop mode is a free fall; The second drop mode is a drop at an angle to the horizontal plane. The third drop mode is a destructive test in which a heavy object is crushed by the falling object; Each time the position is changed, the parameters of the electric energy meter are re-measured based on the standard meter comparison method using the method in step 2, and the parameter information of the re-measured electric energy meter is recorded; The height points for changing the electricity meter include 300 cm, 700 cm, 110 cm and 450 cm; The angles include 30 degrees, 60 degrees and 90 degrees.

4. A method for analyzing measurement errors in an electric energy meter drop test according to claim 1, characterized in that: In the step 4, The electric energy meter collects parameters by processing the measured voltage, current, power, power factor and harmonic parameter values ​​through interquartile range; wherein: The steps of removing outliers by using the interquartile range are as follows: A: The collected data is arranged in ascending order of eigenvalues ​​and divided into three dynamic windows based on the drop test impact time sequence: the pre-drop stability period, the impact transient period, and the post-drop recovery period. Different weights are assigned to the data in each window. B: Calculate the first quartile Q1 and the third quartile Q3 in each dynamic window, where Q1 is the 25th percentile value in the window and Q3 is the 75th percentile value in the window. Use a sliding average correction to eliminate the impact of data fluctuations at the edge of the window. C: Calculate dynamic interquartile range I QR =Q3-Q1, and introduce an adaptive threshold coefficient k, the k value is dynamically adjusted with the intensity of window data fluctuation; D: Remove the window with the size smaller than Q1-k×I QR The sum is greater than Q3+k×I QR Severe outliers are detected, critical outliers are graded and marked, and their corresponding drop shock parameters are retained for subsequent error tracing; E: Establish an abnormal data information association database, and associate the removed severe outliers and marked critical outliers with the impact intensity, duration and other parameters of the drop test.

5. The method for analyzing measurement errors in a drop test of an electric energy meter according to claim 1, characterized in that: In step 5, the method of digital twin-driven dynamic damage modeling and multi-source data fusion is: Step 51: Multi-dimensional data preprocessing and twin initialization Collect high-precision multi-dimensional physical impact data from step 2 and dynamic damage information from step 4, and synchronize the physical and visual data using a spatiotemporal alignment algorithm. Use the preprocessed data as input to construct a digital twin of the electricity meter, including its internal structure, material properties, and boundary conditions, and initialize the mechanical parameters and geometric model of the virtual scene. Step 52: Explicit dynamics simulation and implicit damage visualization prediction Based on the explicit dynamics algorithm, the digital twin is driven to reproduce the drop impact process, dynamically outputting the stress / strain distribution cloud map, displacement trajectory, and material yield threshold exceeding area of ​​the key structure. The damage degree is calculated by the damage accumulation factor, where the damage accumulation factor calculation function is: In formula (1), is the instantaneous stress, is the yield strength of the material, is the stress duration; Step 53: Error analysis using the error correlation model of multi-source data fusion A "damage-error" mapping database was constructed, and the measurement error correlation calculation was performed between high-risk damage areas and corresponding parameters. The multivariate linear regression model was optimized, and the damage factor was introduced as an independent variable. The model formula is: In formula (2), H is the drop height, A is the angle, N is the number of times, T is the temperature, M is the humidity, DCF is the damage accumulation factor, is the regression coefficient, is the error term; Step 54: Dynamically correct the test error The measured metrological errors of the physical prototype are fed back to the digital twin, and simulation parameters are adjusted using an adaptive correction algorithm, resulting in a match of over 90% between the twin's predicted damage areas and the actual test results. Virtual iterative tests are then conducted based on the corrected model to simulate damage evolution and error trends under different drop conditions. Step 55: Damage tracing and error prediction throughout the entire life cycle A damage traceability knowledge base is constructed, and the twin simulation data, multi-source fusion model parameters and actual error values ​​of each test are stored in the blockchain to form an unalterable test archive. Historical data is associated through knowledge graph technology. When new drop parameters are input, the damage-error model of similar working conditions is quickly called to achieve early prediction of measurement errors and locate the key influencing factors of potential damage.

6. A method for analyzing measurement errors in an electric energy meter drop test according to claim 1, characterized in that: The calculation steps of the harmonic analysis method in step 6 are: Step 61: Dynamic time domain signal acquisition and preprocessing The voltage (u(t)) and current (i(t)) signals containing fundamental and harmonic components during the electric energy meter drop test are collected, with the sampling frequency set to 256 times the fundamental frequency to meet the Nyquist criterion. Based on the impact transient period output by the digital twin in step 52, the time domain signals are preprocessed by segmented weighting. Step 62: Improved Fourier Transform and Harmonic Parameter Separation The pre-processed voltage and current signals are converted from time domain to frequency domain using windowed short-time Fourier transform. The Hanning window is used to suppress spectrum leakage, and the window length is dynamically adjusted according to the harmonic order. The transformation formula is: In formula (3), is the Hanning window function, is the correction coefficient of the drop height to the spectrum amplitude, is the correction coefficient of temperature to phase shift, h is the drop height, and T is the ambient temperature; Step 63: Calculation of harmonic power by multi-parameter coupling A harmonic power dynamic correction model is constructed based on the damage accumulation factor output by the digital twin. The calculation formula is: In formula (4), a new damage correction term is added to quantify the impact of hidden damage on harmonic transmission; a and b are the impact of drop height (h) and temperature (T) on harmonic power coefficient respectively; Step 64: Calculation and traceability of layered measurement error The harmonic layered weighted method is used to calculate the energy metering error. The weight is assigned according to the contribution of each harmonic to the total error. The error calculation formula is: In formula (5), is the kth harmonic weight, is the measurement period, is the theoretical electric energy value calculated based on harmonic power during this period, The electric energy meter readings at time t1 and t2 respectively; Step 65: Dynamic calibration of error results The calculated harmonic measurement error is fused and calibrated with the error prediction value output by the multivariate linear regression model in step 53, and the error result is iteratively optimized using the Kalman filter algorithm: the result of the harmonic analysis method is used as the observation value, the result of the regression model is used as the prediction value, the weights of the two are dynamically adjusted, and the calibrated measurement error value is output.

7. A method for analyzing measurement errors in an electric energy meter drop test according to claim 1, characterized in that: In step 7, the random forest algorithm works as follows: Step 71: Physical correlation feature extraction and fusion, extracting multi-dimensional core features and constructing correlation mapping Extract the internal circuit structure parameters, component electrical characteristics, mechanical part dimensions, and time-varying measurement value variations of the electric energy meter; bind the circuit structure parameters to the damage area output by the digital twin; Step 72: Stratified Sampling and Physically Perceptual Feature Selection Damage stratified sampling was used to stratify the original data according to the damage level in the drop test. Bootstrap sampling with replacement was performed within each stratum to ensure that the sample proportions of each damage level were consistent with the actual working conditions. A physical correlation screening mechanism was introduced when randomly selecting features, and split features were preferentially selected from the feature subset that was strongly correlated with measurement error. Step 73: Dynamically constrained decision tree growth and integration The decision tree growth adopts a dual-rule splitting mechanism: minimizing the Gini coefficient is the basic criterion, while physical constraints are introduced. When a split node involves a key parameter, the parameter is mandatory as the basis for splitting, ensuring that the decision tree growth conforms to the laws of engineering physics. 100 independent decision trees are generated according to the predetermined requirements, each tree is grown based on a different sample and feature subset, and the final integration forms a random forest model. Step 74: Multi-type weighted voting prediction and error type determination Input new sample data to each decision tree for parallel prediction: use mean integration for measurement error values, and adopt weighted voting mechanism for error types, assigning weights to each decision tree according to its prediction accuracy in historical data; based on the error type counting formula: In formula (6), T is the prediction error type, is the indicator function, is the output of the j-th tree for the i-th type of error, For the actual prediction results of the jth tree, filter the error type with the highest votes and output the confidence level of the type; Step 75: Cross-condition verification and dynamic model optimization Verification is performed using electricity meter data under different test conditions, and the error prediction deviation under each condition is calculated. When the deviation is greater than 5%, model optimization is triggered, and the decision tree splitting rules are adjusted by adding samples for retraining under the corresponding conditions. An adaptation table is constructed based on the verification results to achieve accurate identification of the differences in the impact of metering errors under different conditions.

Images