On-line verification and diagnosis method for metering box electric energy meter based on multi-source measurement data fusion

Abstract

The application provides a metering box electric energy meter online verification and diagnosis method based on multi-source measurement data fusion, relates to the electric energy metering test technical field, and is executed by an edge Internet gateway: through acquisition of voltage and current sampling sequences and environmental temperature characteristics, time stamp alignment and proportional conversion are executed to output time-synchronous electric quantity characteristic sequences; a group electric quantity fluctuation mean value curve is extracted, and tolerance mapping correction is carried out based on the temperature characteristics to construct a group reference baseline; a to-be-measured sequence is compared with the baseline to solve real-time metering error parameters; based on error parameter and legal error interval comparison, an instruction is generated to trigger the numerical update of the internal metering pulse constant of the electric energy meter, and a state diagnosis label is output. Data asynchronous misplacement and environmental temperature drift interference are effectively overcome, online physical correction and disaster abnormality early warning under non-power-off working conditions are realized, and the stability of diagnosis is enhanced.

Description

Technical Field

[0001] This invention relates to the field of electrical energy metering and testing technology, specifically to an online verification and diagnostic method for metering box electrical energy meters based on multi-source measurement data fusion. Background Technology

[0002] As low-voltage distribution networks evolve towards deep digitalization and intelligence, the reliability of electricity metering equipment directly impacts the transparency of grid operation and the fairness of asset settlement. Traditional online verification and diagnostic technologies for electricity meters are gradually shifting from single, periodic manual inspections relying on external standard instruments to non-intrusive automated sensing based on massive edge data. However, online verification technology still faces significant limitations in complex engineering applications. For example, the existing patent with publication number CN112782638B constructs a fault table through simulation experiments and switches to a backup electricity meter based on abnormal parameter matching. This approach heavily relies on pre-defined destructive fault characteristics and still requires physical switching actions, failing to achieve truly routine, uninterrupted online verification. In real-world edge computing scenarios, the physical nodes within the metering box are constrained by the objective existence of network communication latency. Multi-source sensing data exhibits asynchronous and misaligned characteristics at the transmission layer, resulting in a lack of consistent time reference for subsequent analysis. Outdoor metering boxes are subjected to the interplay of physical stress from long-term exposure to external weather conditions and heavy-load operation. Traditional comparison strategies based on static curves cannot distinguish between "reasonable environmental thermodynamic temperature drift" and "real aging of metering components" in terms of physical logic, and are prone to misjudging common-mode environmental interference as hardware failure, thus generating false positives. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an online verification and diagnostic method for metering box energy meters based on multi-source measurement data fusion. Using the metering box's edge IoT gateway as the core processing hub, it performs benchmark alignment and misalignment removal on the collected data at the time domain. By introducing the group arithmetic mean and ambient temperature characteristics, a dynamically compensated group game baseline is constructed, thereby achieving spatiotemporal decoupling between environmental common-mode interference and hardware intrinsic characteristics within the algorithm space. This solves the problems mentioned in the background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] An online verification and diagnostic method for electricity meters in metering boxes based on multi-source measurement data fusion, executed by the processor of the edge IoT gateway of the metering box, includes the following steps:

[0006] S1: Obtain the multi-channel voltage and current sampling sequence and the corresponding ambient temperature feature data reported by the metering box acquisition node, and perform timestamp reference alignment and reference ratio conversion on the multi-channel voltage and current sampling sequence to output the time-synchronized power feature sequence.

[0007] S2: Perform an arithmetic mean calculation on all time-synchronized power characteristic sequences under the same topological association to extract the group power fluctuation mean curve, and perform tolerance mapping correction on the group power fluctuation mean curve based on the ambient temperature characteristic data to construct a temperature-compensated group reference baseline;

[0008] The time-synchronized energy characteristic sequence of the energy meter under test is compared with the time-series curve value of the temperature-compensated group reference baseline to calculate the real-time metering error parameter characterizing its actual measurement deviation.

[0009] S3: Perform a size comparison between the actual value of the real-time metering error parameter and the pre-configured legal error value allowable range. In response to the determination that the actual value does not fall within the legal error value allowable range, generate an energy meter error calibration command for the energy meter under test based on the real-time metering error parameter.

[0010] The energy meter error calibration command is configured to trigger the internal metering processing unit of the energy meter under test to perform a numerical update of the metering pulse constant and simultaneously output the corresponding status diagnosis association tag.

[0011] Furthermore, the acquisition of the multi-channel voltage and current sampling sequence reported by the metering box acquisition node and the corresponding ambient temperature characteristic data includes:

[0012] Acquire discrete meteorological temperature sampling sequences transmitted from external meteorological interface nodes;

[0013] By combining the thermal conduction delay parameter of the metering box, a first-order low-pass filtering algorithm is used to perform smooth interpolation processing on the discrete meteorological temperature sampling sequence to generate the ambient temperature feature data that matches the multi-channel voltage and current sampling sequence in terms of time granularity.

[0014] Furthermore, prioritizing the multi-channel voltage and current sampling sequence, timestamp reference alignment and reference scaling are performed to output a time-synchronized power characteristic sequence, including:

[0015] Based on the preset node power conservation rule, the asynchronous sampling time difference in the multi-channel voltage and current sampling sequence is analyzed, and dynamic time window matching processing is performed to remove the waveform misalignment error caused by the transmission delay due to network transmission time lag.

[0016] After removing the waveform misalignment error caused by the transmission delay, the measurement data stream is scaled using a voltage and current unified dimension conversion ratio to extract the time-synchronized power feature sequence that can be directly compared in the same time dimension.

[0017] Furthermore, prioritizing the application of tolerance mapping correction to the mean curve of the group's power fluctuation based on the ambient temperature characteristic data, a temperature-compensated group reference baseline is constructed, including:

[0018] Extract the historical average load rate characteristic data of the energy meter under test;

[0019] In response to the determination that the historical average load rate characteristic data exceeds the preset rated high load heating threshold, a heat penalty attenuation weight is assigned to the energy meter under test based on the ambient temperature characteristic data to block the benchmark calculation deviation caused by the local self-heating of the energy meter under test.

[0020] By combining the preset equivalent coefficient of temperature rise of meter components, the ambient temperature characteristic data is mapped to a global temperature compensation weight value.

[0021] The global temperature compensation weight value and the heat penalty attenuation weight are multiplied and weighted. The weighted result is used to perform dynamic proportional stretching calculation on the upper and lower limit fluctuation tolerance range of the average fluctuation curve of the group power fluctuation. In response to the determination that the stretched tolerance boundary exceeds the hardware operation safety extreme value threshold, the tolerance cut-off protection logic is triggered to forcibly solidify the current boundary, thereby generating the temperature-compensated group reference baseline that includes the dynamic allowable tolerance range.

[0022] Furthermore, the step of performing a time-synchronized energy characteristic sequence comparison operation between the energy meter under test and the temperature-compensated group reference baseline to calculate the real-time metering error parameter characterizing its actual measurement deviation includes:

[0023] The time-synchronized energy characteristic sequence of the energy meter under test is projected onto the temperature-compensated group reference baseline containing the dynamic allowable tolerance range for time-series curve numerical comparison.

[0024] In response to determining that the time-synchronized power characteristic sequence of the energy meter under test exceeds the boundary of the dynamic allowable tolerance range, the power difference value exceeding the tolerance range in the part exceeding the corresponding boundary is extracted.

[0025] The power difference exceeding the tolerance range is subjected to high-frequency electrical noise smoothing filtering to remove common-mode interference components caused by grid voltage transients. Principal component extraction is used to reduce the dimensionality of the real-time metering error parameter characterizing the physical aging drift of the metering element of the energy meter under test.

[0026] Furthermore, based on the real-time metering error parameters, a meter error calibration command is generated for the meter under test, including:

[0027] The actual value of the real-time measurement error parameter is compared with the pre-configured legal error value allowable range.

[0028] In response to the determination that the actual value of the real-time measurement error parameter does not fall within the legally allowed error value range, the static error offset of the real-time measurement error parameter is extracted, and the measurement pulse constant compensation factor is extracted based on the multiplication of the static error offset with the reference pulse constant of the current instrument under test and the rounding operation of truncating the decimal places.

[0029] The metering pulse constant compensation factor is encapsulated in a downlink communication control message to generate the energy meter error calibration command, thereby driving the internal metering processing unit of the energy meter under test to load the metering pulse constant compensation factor and overwrite the metering pulse constant of the current measurement cycle.

[0030] Furthermore, the synchronous output corresponding to the state diagnosis association tag includes:

[0031] Extract multiple real-time measurement error parameters calculated within multiple consecutive historical detection cycles to construct a historical error record set;

[0032] Perform error change trend slope fitting calculation on the historical error record set to generate error change slope value characterizing the physical aging evolution rate of the internal measuring components of the energy meter under test;

[0033] In response to the determination that the slope value of the error change exceeds the preset safety degradation threshold slope, an on-site inspection and maintenance instruction characterizing the aging and degradation trend of the component is generated.

[0034] The on-site inspection and maintenance instructions are output as the status diagnosis association tags, which are configured to trigger the operation and maintenance scheduling server to generate inspection instructions for the energy meter under test.

[0035] Furthermore, the trigger tolerance truncation protection logic to enforce the current boundary includes:

[0036] For the state where the dynamic allowable tolerance boundary after stretching exceeds the hardware operation safety extreme value threshold, a time accumulation statistical operation is performed to extract the over-limit duration statistic used to characterize the degree of high temperature over-limit;

[0037] The above-limit duration statistic is compared with the pre-configured hardware thermal overload tolerance time constant.

[0038] In response to determining that the over-limit duration statistic is less than or equal to the hardware thermal overload tolerance time constant, the static extreme value locking execution logic is activated; the locally configured steady-state cutoff benchmark threshold is retrieved as the absolute constraint extreme value, and the absolute constraint extreme value is used to perform a constant assignment operation on the dynamic allowable tolerance boundary, locking it as a fixed extreme value constant;

[0039] In response to the determination that the over-limit duration statistic is greater than the hardware thermal overload tolerance time constant, the tolerance boundary dynamic expansion control logic is triggered; based on the division of the over-limit duration statistic with the preset step time constant and the natural logarithm extraction operation, the time-varying boundary relaxation correction factor is calculated.

[0040] The time-varying boundary relaxation correction factor and the steady-state truncation benchmark threshold are calculated by numerical scaling multiplication to dynamically generate the relaxed compensation tolerance boundary value.

[0041] The relaxed compensation tolerance boundary value is used to replace and update the stretched dynamic allowable tolerance boundary, so that the updated boundary is used as the current configuration boundary to participate in the time series curve numerical comparison calculation in subsequent cycles, in order to avoid the system continuously triggering invalid over-limit interception misjudgments under long-term abnormal temperature conditions.

[0042] Furthermore, the real-time metering error parameter characterizing the physical aging drift of the metering element of the energy meter under test is obtained by principal component extraction, including:

[0043] Obtain the power difference sequence that exceeds the tolerance range for all associated nodes under the same topology association, perform feature data array splicing operation on multiple power difference sequences, and construct a multi-node error covariance matrix that characterizes the comprehensive error distribution characteristics of the equipment group within the same power supply topology;

[0044] Perform orthogonal decomposition calculation on the multi-node error covariance matrix to extract the principal component feature vector set representing different error evolution directions;

[0045] Obtain the sequence of power difference values ​​exceeding the tolerance range corresponding to the power meter under test, project it onto the principal component feature vector set, and calculate the Euclidean distance deviation feature parameter of the sequence relative to the mean center of error fluctuation of the equipment group in the distribution area after projection;

[0046] The principal component feature vector set is subjected to a data separation and judgment operation using a preset feature distance deviation comparison threshold.

[0047] In response to determining that the Euclidean distance deviation feature parameter is less than or equal to the corresponding distance threshold, the feature component is identified as a common-mode temperature drift error component caused by external meteorological temperature change, and is filtered out from the subsequent individual error evaluation calculation.

[0048] In response to the determination that the Euclidean distance deviation feature parameter is greater than the corresponding distance threshold, feature components that present an isolated distribution state are filtered out and identified as the intrinsic aging deviation vectors of individual components that have been excluded from the interference of group environmental temperature drift.

[0049] The intrinsic aging deviation vector of the individual component is extracted as the target physical quantization error feature parameter. The absolute value of the geometric modulus of the feature vector in space is calculated, and the absolute value of the geometric modulus is converted into a one-dimensional scalar value based on a preset full-scale reference, and output as the real-time measurement error parameter.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This invention outputs a time-synchronized power characteristic sequence by performing timestamp benchmark alignment and reference scaling on multi-channel voltage and current sampling sequences. This eliminates waveform misalignment errors and physical dimension differences caused by edge network transmission delays, enabling discrete asynchronous multi-source data to have absolute algebraic comparability in the same spatiotemporal dimension, laying a reliable input benchmark for subsequent high-precision feature comparison. By extracting the average curve of group power fluctuations and performing tolerance mapping correction based on ambient temperature characteristic data, a temperature-compensated dynamic group reference baseline is constructed. This mechanism utilizes the collaborative characteristics of multi-node group data to achieve deep spatiotemporal decoupling between reasonable ambient temperature drift and the actual physical aging drift of measuring components, effectively suppressing false positive alarms under meteorological conditions in traditional static threshold schemes, and improving the sensitivity and stability of fault identification under complex operating conditions.

[0052] Based on a closed-loop comparison of real-time metering error parameters with the legally permissible error range, a calibration command is generated to directly trigger the internal execution of the metering pulse constant overwriting and updating, and a status diagnostic associated tag is output simultaneously. This effectively reduces the frequency of power grid service interruptions and enables joint processing of real-time closed-loop control of metering accuracy and operation and maintenance scheduling. Attached Figure Description

[0053] Figure 1 This is a full-link system architecture diagram and data flow diagram of the online verification and diagnosis method for metering boxes and energy meters based on multi-source measurement data fusion;

[0054] Figure 2 This is a schematic diagram of a scenario for multi-source heterogeneous spatiotemporal feature fusion and deasynchronous preprocessing provided by the present invention;

[0055] Figure 3 This is a schematic diagram of the present invention based on physical intrinsic space dimensionality reduction decoupling and multidimensional diagnostic label output;

[0056] Figure 4This is a flowchart of the data spatiotemporal alignment and tolerance boundary dynamic expansion control of the present invention;

[0057] Figure 5 This is a schematic diagram of the calculation results of the numerical verification experiment conducted on the dynamic expansion control characteristics of the tolerance boundary according to the present invention.

[0058] Figure 6 This is a flowchart of the multidimensional error feature dimensionality reduction decoupling and dual-track diagnosis closed-loop execution process of the present invention. Detailed Implementation

[0059] The technical solutions in 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0060] Example 1:

[0061] Please see Figures 1 to 6 The present invention provides a technical solution:

[0062] An online verification and diagnostic method for electricity meters in metering boxes based on multi-source measurement data fusion, executed by the processor of the edge IoT gateway of the metering box, includes the following steps:

[0063] Step S1 involves entering the zero-time data acquisition phase. This includes acquiring multi-channel voltage and current sampling sequences reported at millisecond-level high-frequency cycles by intelligent sensing physical acquisition nodes deployed within the metering box (e.g., built-in high-frequency current transformers and high-frequency ADC conversion channels). Simultaneously, it acquires discrete meteorological temperature sampling sequences periodically transmitted hourly by external meteorological interface nodes via a wide-area IoT communication link.

[0064] Define the thermal conduction delay parameter of the enclosure. Specifically, the values ​​are determined by constructing a controlled thermodynamic data offline playback sandbox, continuously injecting discrete meteorological temperature mutation sequences from historical records into it; continuously monitoring the virtual internal temperature fluctuation patterns output by the corresponding first-order low-pass filter model inside the sandbox, and extracting the distortion error caused by its temporal misalignment with the high-frequency electrical sampling sequence. A physical constraint rule is used to ensure that the false positive following rate of the high-frequency electrical waveform and low-frequency meteorological data converges to below 1% on the time axis. The heat conduction delay parameter of the chamber is extracted and solidified through time window tolerance. It represents the hysteresis physical time constant that indicates the penetration of external meteorological temperature changes into the physical material of the sealed metering box, thereby causing the internal electronic measuring components to generate a physical temperature rise at the same frequency. In this preferred embodiment, its preferred value is defined as a time offset range of 1800 seconds to 3600 seconds.

[0065] Combining the thermal conduction delay parameters of the enclosure A first-order low-pass filtering algorithm is used to perform smooth interpolation on discrete meteorological temperature sampling sequences, reconstructing the discrete, step-like low-frequency meteorological data into continuous, smooth high-frequency curves. This generates environmental temperature feature data that is perfectly matched and aligned with the multi-channel voltage and current sampling sequences in terms of time granularity. Specifically, this is achieved by retrieving the baseline environmental temperature value from the previous sampling time node, denoted as... And obtain the discrete meteorological temperature sampling value input from the outside at the current time, denoted as Construct the following smooth interpolation calculation logic: It samples discrete meteorological temperature values ​​at the current moment. Subtract the baseline ambient temperature value of the previous node Obtain the real-time temperature difference; extract the absolute time difference between two consecutive sampling actions. and the time difference Divide by the thermal conduction delay parameter of the enclosure The dimensionless smooth attenuation multiplier is obtained; the real-time temperature difference is multiplied by this smooth attenuation multiplier to obtain the compensated temperature variable; this compensated temperature variable is then multiplied by the reference ambient temperature value of the previous node. Add them together and output the ambient temperature characteristic data at the current moment. .

[0066] Further, the preset node power conservation rule is extracted. Specifically, by acquiring the physical node data of the current distribution substation, the actual total power flowing into the substation is extracted. Simultaneously, the power data of each outflow branch and a fixed physical line loss heating value are acquired. The power data of each outflow branch and the physical line loss heating value are added together to obtain the total output value. A numerical equality balance constraint condition is constructed to ensure that the total inflow power data and the total output value are equal. Based on this numerical equality balance constraint condition, the sampling time window of the multi-channel voltage and current sampling sequence is shifted until the shifted sampling features satisfy the above balance condition of equal inflow and outflow values, thereby extracting the asynchronous sampling time difference features in the multi-channel voltage and current sampling sequence. Using the node power conservation rule as a convergence constraint condition, a discrete-time sliding correlation calculation model is constructed for voltage and current sampling data with asynchronous misalignment characteristics to perform dynamic time window matching processing. Specifically, the sampling feature sequence of the upstream reference topology node is set as follows: The current target measurement point sampling feature sequence to be aligned is set as follows: Introduce a time offset variable representing the relative delay. Furthermore, a discrete cross-correlation function equation is established for extreme value matching to dynamically shift the sampling window of the measurement data along the time axis: ;

[0067] Control time offset variable Within a preset legal network delay interval, a discrete sliding operation is performed point by point, and the corresponding cross-correlation values ​​are output synchronously. The two electrical data streams are considered to have the highest overlap on the physical waveform if and only if the cross-correlation value reaches the global maximum value. The corresponding time offset variable is then extracted. The numerical value serves as the final waveform misalignment compensation value. This time offset variable... This refers to the actual transmission delay deviation between asynchronous network communication nodes. In this embodiment, based on the physical bandwidth congestion characteristics of the narrowband IoT in the distribution area, the actual sliding search range is preferably defined within the delay constraint interval of [-50 milliseconds, 50 milliseconds]. The parameter N represents the total number of discrete sampling points within a single operation window; in this embodiment, it is preferably configured as 256 standard data points within a single power frequency waveform period. This actively removes and forcibly eliminates waveform misalignment errors caused by the concurrent transmission delay of the edge network.

[0068] This embodiment further verifies the alignment effectiveness of the above discrete-time sliding correlation calculation model in the physical network. Based on the offline standard dataset (extracted from the physical packet capture measurement logs of low-voltage distribution substations under different network congestion levels), time window matching processing is performed on multiple asynchronous voltage and current sequences. The results of extreme value optimization matching and time difference stripping calculation are shown in the table below:

[0069] Table 1: Results of Asynchronous Sampling Time Difference Extremum Optimization Matching Calculation under Offline Standard Dataset

[0070] Network congestion status label Number of discrete points (N) in the sampled feature sequence Initial physical time waveform misalignment error (milliseconds) Discrete cross-correlation numerical (R(τ)) extreme values Match the output time offset variable (milliseconds) Network transmission delay error reduction (%) Idle networks are common 256 5 0.985 5 98.50% Mild concurrent congestion 256 18 0.972 18 97.20% Severe sudden traffic congestion 256 45 0.958 45 95.80% Extreme Limit Communication 256 62 0.82 50 (Extreme Limit) 80.60%

[0071] Table 1 shows the reduction in network transmission time delay error. This is achieved by extracting the absolute value of the initial physical time waveform misalignment error, subtracting the absolute difference between this misalignment error and the time offset variable of the matched output, and obtaining the effective compensation amount. This effective compensation amount is then divided by the absolute value of the initial physical time waveform misalignment error to obtain the quotient. Finally, this quotient is multiplied by a constant of 100 to convert it into a one-dimensional percentage output. The quantification of network transmission time delay error reduction utilizes power conservation and cross-correlation extreme value optimization logic to determine the effective compensation ratio for forced physical alignment of asynchronous data under different network degradation levels.

[0072] As shown in Table 1, based on calculations and extractions from offline standard datasets, when facing regular and severe network congestion ranging from 5 milliseconds to 45 milliseconds, the discrete cross-correlation values ​​can clearly pinpoint a single maximum value (0.985, 0.972, and 0.958, respectively), and the calculated time offset variable can cover the initial time delay, with an error reduction of over 95%. However, when the network exhibits extreme communication exceeding the limit of 62 milliseconds, due to exceeding the locally set delay constraint range of [-50 milliseconds, 50 milliseconds], an extreme value limiting action is executed, forcing an output boundary value of 50 milliseconds.

[0073] Furthermore, this embodiment introduces a unified dimension conversion ratio for voltage and current, denoted as . Specifically, the linear scaling formula is constructed for the heterogeneous dimensional features in the measurement data stream as follows: It obtains the original feature sequence after removing the transmission delay waveform misalignment error; Convert it to a unitary ratio with the pre-configured voltage and current. Scalar multiplication dimensionality reduction is performed to eliminate unit physical attributes, thereby mapping both onto the same dimensionless computational plane. This extracts time-synchronized electrical characteristic sequences that are within the same dimensionless time dimension and possess absolute algebraic comparability. The conversion ratio for voltage and current to a unified dimension is as follows. In this preferred embodiment, the preferred value is set to a normalization range of 0.01 to 0.05. This ensures that the voltage and current characteristics will not be distorted in the subsequent numerical calculation due to the different original dimensional amplitudes when constructing the high-dimensional error covariance matrix.

[0074] Specific implementation details for step S2: Performing group baseline game theory, time-sensitive asymmetric compromise, and dimensionality-reduced topology divergence stripping: Specifically, this involves obtaining the time-synchronized power characteristic sequences of all target nodes under the same physical power supply topology association. A horizontal arithmetic mean is calculated on all time-synchronized power characteristic sequences under the same topology association to filter out random jump interference from individual devices, extracting the group power fluctuation mean curve that reflects the actual noise floor fluctuation of the current distribution area's equipment group.

[0075] During the group baseline game phase, the control module extracts the historical average load rate characteristic data of the energy meter under test within the past rolling evaluation period. In this embodiment, the past rolling evaluation period is preferably configured as a rolling time window of 24 consecutive hours or 7 days. The historical average load rate characteristic data is established through the following algebraic rules: the control gateway retrieves all historical load current sampling points recorded by the energy meter under test within the rolling time window, calculates the arithmetic average operating current value of all sampling points, and divides this average operating current value by the rated maximum allowable operating current of the energy meter, thereby extracting a one-dimensional percentage scalar value representing the overall load level within the [0,1] interval. The rated high load heating threshold is initialized and pre-configured, denoted as... This threshold defines the boundary between the physical interference of high-current Joule heat from the meter's own components and the dominance of environmental climate heat. This threshold resides in the global state registry as a static reference point for binary state triggering. Its dynamic reading and state transition conditions are as follows: In response to the input of historical average load rate characteristic data, the control flow performs a numerical boundary condition judgment and comparison operation; when it is determined that the current load characteristic has not crossed this boundary and is in a normal low-power sleep logic state, the current environmental benchmark calculation state is maintained and allowed to proceed; in response to the determination that the historical average load rate characteristic data exceeds the boundary and is greater than the rated high-load heat generation threshold... When the system is in the overheating activation logic state, a control command to mitigate thermal contamination is immediately triggered and executed. In this embodiment, the rated high-load heating threshold is... The quantization value is fixed at 85% of the device's rated maximum allowable operating current. If the rated high load heating threshold is set too low, it is easy to misjudge the reasonable high load data flow during the normal summer evening peak period as a serious heating abnormality, causing excessive false positive attenuation intervention; if the rated high load heating threshold is set too high, there is a risk of masking the hidden danger of burnout caused by poor contact of electrical terminals.

[0076] Introducing a heat penalty decay weight, denoted as This parameter obtains the internal maximum temperature characteristic data in real time from the temperature probe deployed on the physical casing of the energy meter under test through the physical bus interface, and materializes it into a local single-point maximum temperature value; extracts the single-point maximum temperature value collected by the temperature probe, and subtracts the meteorological reference temperature value issued by the weather station at the same time to obtain the temperature difference between the two; obtains the pre-fixed maximum allowable temperature difference limit of the device at 40°C, divides the previously calculated temperature difference value by the maximum allowable temperature difference limit of the device to obtain a basic ratio; the direct difference obtained by subtracting the basic ratio from the basic constant 1 is the heat penalty attenuation weight. In this embodiment, the preferred value is defined as a closed range of 0.2 to 0.5. If the heat penalty attenuation weight is too small, it may easily negate the temperature compensation response of the heating device, causing a break in the temperature correlation mapping feature chain. The heat penalty attenuation weight is then assigned to the energy meter under test based on ambient temperature characteristic data. This is to proactively prevent the deviation in baseline calculations caused by localized self-heating pollution of environmental parameters.

[0077] Read the equivalent coefficient of temperature rise of the meter components, and record it as: This parameter represents the linear drift mapping ratio of the physical impedance characteristics of the internal semiconductor metering components to each degree Celsius change in external temperature. In this embodiment, a standard room temperature of 20°C is used as the reference zero point. The absolute value of the temperature difference between the ambient temperature characteristic data and this reference zero point is extracted, and this absolute value of the temperature difference is compared with the equivalent coefficient of the meter component's temperature rise. A scalar multiplication operation is performed, and the output value is added to the base constant 1 to construct the global temperature compensation weight value. In this embodiment, the equivalent coefficient for the temperature rise of the meter components... The value is fixed at 0.05% / ℃. If the equivalent coefficient of temperature rise of the meter components is set larger, it is easier to over-amplify slight natural temperature differences, resulting in an abnormally high global temperature compensation weight value, which makes normal power fluctuations mistakenly identified as abnormal data caused by environmental pollution; if the equivalent coefficient of temperature rise of the meter components is set smaller, it is easier to be sluggish in responding to semiconductor drift in real extremely cold or hot environments, thus losing the meaning of compensation.

[0078] This embodiment establishes the initial baseline width of the population power fluctuation mean curve before performing dynamic proportional stretching. Historical discrete measurement point data of the population power fluctuation mean curve under historical standard room temperature conditions (fixed within a 20℃ range) are extracted during continuous, stable operation cycles without abnormalities. For these discrete measurement points, the absolute discrete difference from the central mean is calculated, and the maximum value in this discrete difference sequence is extracted as the basic one-sided tolerance value. This basic one-sided tolerance value is added to and subtracted from the mean curve to construct the initial upper and lower limit fluctuation tolerance range of the population power fluctuation mean curve. The global temperature compensation weight value and the heat penalty attenuation weight are then used... Perform a numerical proportional multiplication calculation, and use the dynamic amplification multiplier obtained after multiplication to perform a dynamic proportional stretching operation by multiplying the initial upper and lower limit fluctuation tolerance range constructed above, thereby generating a dynamic allowable tolerance boundary that is linked to temperature.

[0079] Initialize the hardware operation security threshold in the global registry, denoted as This parameter serves as a hard state transition limit for triggering the tolerance cutoff protection logic, representing the factory-set upper limit of environmental high-temperature drift tolerance for the metering equipment. In this preferred embodiment, this threshold is configured as three times the extreme value of the legal standard tolerance (±2%). A smaller hardware operation safety threshold setting makes it easier for abnormal features to cascade and spread, leading to full-link degradation. This can cause reasonable physical temperature drift caused by normal, short-term high temperatures in summer to be misjudged as equipment damage, triggering widespread false alarms and resource congestion across the network. Conversely, a larger hardware operation safety threshold makes it easier to lose monitoring and perception of the actual hardware burn-out threshold, causing real physical fault data to be transmitted normally between upstream and downstream modules.

[0080] For dynamic tolerance boundaries exceeding the hardware operating safety threshold The abnormal state execution time is accumulated and statistically analyzed to extract the over-limit duration statistic, representing the severity of high-temperature physical over-limits. A time window constraint item is independently defined in the configuration library, and a hardware thermal overload tolerance time constant is introduced, denoted as... This parameter serves as the distinguishing point between the two state machine mechanisms of "single-point overload circuit breaker" and "macroclimate compromise," representing the maximum tolerable time span for safe residence within the edge of the over-limit alarm zone. In this preferred embodiment, it is set to 72 consecutive hours.

[0081] Combine the over-limit duration statistic with the hardware thermal overload tolerance time constant Perform a numerical comparison. The determination is made when the out-of-limit duration statistic is less than or equal to the hardware thermal overload tolerance time constant. When a short-term overload fault occurs in a single meter, the static extreme value locking execution logic is activated. A steady-state cutoff reference threshold, pre-stored in the local read-only memory, is retrieved as the absolute constraint extreme value. This steady-state cutoff reference threshold is configured independently of real-time temperature flow as the maximum allowable tolerance point value in the meter design specifications; in this embodiment, it is preferably a fixed error limit of 8%. The control logic node extracts the dynamic allowable tolerance boundary parameter and performs a direct constant overwrite assignment operation on it using the steady-state cutoff reference threshold, forcibly locking it to this fixed extreme value constant for the subsequent calculation lifecycle.

[0082] When the statistical value of the over-limit duration is greater than the hardware thermal overload tolerance time constant. This indicates that the climate environment is in a prolonged state of severe degradation, triggering the dynamic expansion control logic of the tolerance boundary; a time-varying boundary relaxation correction factor is introduced, denoted as... This refers to a floating expansion multiplier that undergoes adaptive boundary compromise to avoid widespread misjudgment and paralysis. Its numerical output upper limit truncation value is set to 1.5 to limit the final relaxed compensation tolerance boundary value from exceeding 1.5 times the steady-state truncation benchmark threshold. Specifically, this is achieved by obtaining the out-of-limit duration statistics (denoted as...). The time-varying boundary relaxation correction factor is generated directly through sequential algebraic operations. The algebraic operation is presented as follows: It achieves this by sequentially performing the following arithmetic steps: extracting the over-limit duration value... Subtract the hardware thermal overload tolerance time constant value Calculate the difference value exceeding the tolerance time; divide this difference value by the preset smoothing step time constant value. (In this embodiment, 24 hours is preferred), to obtain a time quotient; add a base constant of 1 to the time quotient and extract the natural logarithm of the summed result; then combine the natural logarithm with a fixed compensation amplification factor. (In this embodiment, 0.15 is preferred) Perform a numerical multiplication operation and add the basic constant 1 to the product result to finally obtain the time-varying boundary relaxation correction factor. When the value calculated using the logarithmic formula exceeds 1.5, the control flow performs hard limiting interception, and the time-varying boundary relaxation correction factor is applied. The maximum output value is locked at 1.5. This compensation amplification factor... The smaller the value, the more likely it is to cause a lag in tolerance relaxation, making it difficult to promptly alleviate the global verification paralysis caused by macro-meteorological anomalies; the generated time-varying boundary relaxation correction factor The calculation is performed by numerical scaling multiplication with the steady-state truncation reference threshold. The direct numerical multiplication operation is to apply the calculated time-varying boundary relaxation correction factor. The value is directly multiplied by the value of the steady-state cutoff benchmark threshold, and the product of the two is directly assigned to update the relaxed compensation tolerance boundary value.

[0083] The current boundary configuration is replaced and updated using the relaxed compensation tolerance boundary value, thereby preventing global logic check lock-up and crash caused by severe weather stay, and generating the final temperature-compensated group reference baseline.

[0084] Furthermore, the time-synchronized energy characteristic sequence of the energy meter under test is projected onto a group reference baseline that includes relaxed compensation tolerance boundary values, and a time-series curve numerical comparison operation is performed.

[0085] When the time-synchronized power characteristic sequence exceeds the corresponding dynamic allowable tolerance boundary, extract the power difference sequence of the portion exceeding the tolerance range.

[0086] High-frequency electrical noise smoothing filtering is performed on the power difference sequence exceeding the tolerance range. Specifically, the current discrete electrical sampling frequency of the data acquisition node is obtained, and the sampling frequency value is divided by 2 to extract the basic Nyquist frequency reference. The empirical low-frequency baseline value (defined as a 500Hz parameter in this embodiment) of transient common-mode interference generated by the start-up and shutdown of large nonlinear load equipment in the power grid is obtained by consulting the pre-configured hardware characteristic dictionary. The pre-configured hardware characteristic dictionary is represented as a structured key-value pair configuration table (Key-Value-Map) in the local read-only storage area of ​​the edge IoT gateway. It is configured with corresponding empirical low-frequency baseline values ​​according to the typical nonlinear equipment types commonly found in the distribution substation where the metering box is located. For example, when the equipment type key value is a group of conventional variable frequency air conditioners, the mapped low-frequency baseline value is 350Hz; when the equipment type key value is a high-power inductive motor, the mapped low-frequency baseline value is 500Hz. During the initialization phase, the physical environment tag of the currently deployed substation is read, and then the corresponding empirical baseline frequency is automatically retrieved from the dictionary to participate in subsequent calculations. The Fast Fourier Transform (FFT) algorithm is further invoked to map the input sequence in the one-dimensional time domain state into a complex frequency domain feature matrix to deconstruct the feature amplitude and phase information of each dimension. Using the aforementioned 500Hz value as a constraint, the feature amplitude frequency band of the frequency domain feature matrix in the range of 500Hz to Nyquist frequency is directly zeroed out. The Inverse Fast Fourier Transform (IFFT) algorithm is executed on the frequency domain feature matrix after suppression and zeroing to remap and restore it to a smooth electrical quantity sequence in the time domain dimension. Utilizing the low-pass frequency cutoff characteristic with a clear physical numerical boundary, the high-frequency transient common-mode interference components caused by equipment start-up and shutdown are deterministically eliminated.

[0087] Further, obtain the power difference sequence corresponding to all associated nodes under the same power supply topology that exceeds the tolerance range. Perform a feature data array horizontal splicing operation on multiple one-dimensional power difference sequences.

[0088] The feature scheduling unit introduces a multi-node error covariance matrix, denoted as... It uses the device identification code of the upper-level distribution area gateway as the spatial topology association identifier, traverses and retrieves all target concurrent measurement points belonging to the same power supply topology association within the system data bus; extracts the power difference sequence exceeding the tolerance range corresponding to each target node according to a unified time-series sampling window, and uses it as an independent column vector; performs a horizontal array splicing operation on the column vectors of a total of N1 nodes to construct a dimension of The feature array is denoted by K, where K is the number of time-series sampling points in a single sequence. The final multi-node error covariance matrix is ​​output by calculating the covariance between column vectors. The multi-node error covariance matrix is ​​configured to represent the comprehensive error co-fluctuation pattern exhibited by a group of supporting equipment due to their shared external macro-meteorological environment. The constructed multi-node error covariance matrix... Perform orthogonal eigenvalue decomposition based on algebraic characteristic equations. Specifically, based on the linear algebraic space mapping theorem, construct the following fundamental characteristic equations within the computational unit: Further, algebraic solutions are performed on the equation to obtain all non-negative eigenvalues ​​that satisfy the balance constraint, and the eigenbase vectors corresponding to each eigenvalue. In this embodiment, due to the input multi-node error covariance matrix... The data structure exhibits real symmetry, and the calculated feature vector basis vectors of each dimension are orthogonally isolated in spatial topology, directly representing the independent fluctuation dimensions of different physical error evolution directions. These orthogonal feature basis vectors are extracted as the principal component feature vector set after dimensionality reduction. Parameters This refers to the absolute error fluctuation energy weights along each orthogonal dimension; its actual calculated value is a set of discrete real numbers greater than or equal to zero. The parameter V represents the set of discrete eigenvectors that carry this error energy.

[0089] The following matrix truncation feature convergence action is further performed: The full eigenvalue sequence corresponding to the multi-node error covariance matrix is ​​calculated, and all eigenvalues ​​are sorted in descending order of value. The sorted eigenvalues ​​are then accumulated sequentially, and the current accumulated value is divided by the sum of all eigenvalues ​​to calculate the cumulative contribution ratio. In this embodiment, 85% is set as a fixed threshold parameter for dimension truncation. When the cumulative contribution ratio is determined to be greater than or equal to 85%, subsequent accumulation actions are stopped, and the feature components corresponding to all eigenvalues ​​involved in the current accumulation are retained as principal component feature vector sets representing different physical error evolution directions. Simultaneously, the control flow directly clears and discards all remaining feature components not included in the accumulation range. This contribution rate threshold constraint action completes the extraction and filtering of the entity feature set.

[0090] Obtain the sequence of electricity difference values ​​exceeding the tolerance range corresponding to the electricity meter under test, project it onto the principal component feature vector set, and calculate the Euclidean distance deviation feature parameter of this dimensionality-reduced sequence relative to the mean center of the error fluctuation of the equipment group in the distribution area after spatial projection. Specifically, after dimensionality reduction and reconstruction, extract global features and introduce a feature distance deviation comparison threshold, denoted as . This is used to geometrically distinguish between "group-wide common temperature drift" and "individual hardware intrinsic aging" in a reduced-dimensional logical space. It retrieves the full sample point dataset after principal component projection and calculates the geometric scatter standard deviation of the data distribution in this space using statistical variance. The feature distance deviation from the comparison threshold is calculated through pure normalized numerical amplification. The value of is determined to be three times the calculated standard deviation and the absolute distance. If the feature distance deviation from the comparison threshold is smaller, it is easier to treat reasonable meteorological temperature drift as an independent aging bias and separate it, causing false positives and blocking of compliant sample streams;

[0091] Using feature distance deviation to compare thresholds Perform data separation and judgment operations on the principal component eigenvector set.

[0092] When the Euclidean distance deviation feature parameter is determined to be less than or equal to the corresponding feature distance deviation comparison threshold At that time, the feature component is physically identified as a common-mode temperature drift error component caused by large-scale temperature changes in external weather, and the data cleaning logic is triggered to filter and remove it from the subsequent individual residual sequence core evaluation calculation cache.

[0093] When the Euclidean distance deviation from the feature parameter is determined to be greater than the corresponding feature distance deviation comparison threshold, In this process, feature components exhibiting extremely isolated distributions in the reduced-dimensional physical space are selected and identified as the intrinsic aging deviation vectors of individual components, excluding physical interference from group environmental temperature drift. These intrinsic aging deviation vectors of individual components are then extracted as core objective physical quantification error feature parameters. Further dimensional scaling and dimensionality reduction are performed: by extracting the values ​​of each component in the multidimensional feature vector, squaring each component value, and summing them, the square root of the sum is calculated to determine the geometric modulus of the feature vector in Euclidean space, representing its absolute deviation from the physical magnitude; the reference rated full-scale value of the current energy meter under test is retrieved from the local configuration library (e.g., the rated maximum measurable power or energy reference extreme value calibrated by the meter at the factory), the geometric modulus value is divided by the reference rated full-scale value, and the quotient is multiplied by the constant 100, forcibly mapping it from a multidimensional absolute physical quantity to a one-dimensional scalar value with pure percentage dimensions; this one-dimensional scalar value is encapsulated to generate and output as a real-time metering error parameter for the actual measurement degradation degree of the physical device, which can be directly called and compared in subsequent stages.

[0094] Furthermore, this embodiment demonstrates the ability of the multi-node error covariance matrix to decouple and clean common-mode temperature drift and intrinsic aging. Historical operating data (from data collection records of transformer substations experiencing extreme high-temperature weather for 7 consecutive days under the physical power supply topology) was extracted, and orthogonal eigenvalue decomposition and dimensionality reduction distance comparison were performed on branch nodes within the same physical topology within the transformer substation. The specific feature space separation results are shown in the table below:

[0095] Table 2: Results of Dimensionality Reduction and Intrinsic Physical Aging Separation of Multi-Node Error Features from Actual Operation Records of the Transformer Area

[0096] Power supply topology target node identifier Cumulative contribution ratio of all eigenvalues ​​(%) Euclidean distance from the center of the relative fluctuation mean deviation characteristic parameter Feature distance deviation comparison threshold Isolated distribution, separation, qualitative state Real-time measurement error parameter output value (%) Node A01 88.50% 1.25 3 Interception and rejection (common mode temperature drift) 0.00% (Data cleared) Node A02 89.20% 2.1 3 Interception and rejection (common mode temperature drift) 0.00% (Data cleared) Node A03 87.80% 2.85 3 Interception and rejection (common mode temperature drift) 0.00% (Data cleared) Node B01 91.40% 5.42 3 Screening out and retaining (intrinsic aging) 3.15%

[0097] The qualitative state of isolated distribution separation is determined by reading the Euclidean distance deviation feature parameter calculated by dimensionality reduction and comparing it with the feature distance deviation comparison threshold. If the Euclidean distance deviation feature parameter is less than or equal to the feature distance deviation comparison threshold, an "intercept and remove" control label is written to the register; if the Euclidean distance deviation feature parameter is greater than the feature distance deviation comparison threshold, a "screen out and retain" control label is written to the register. This qualitative state directly reflects the logical control action of using geometric divergence boundaries to diagnose and cut off the common-mode temperature drift and intrinsic aging of the target node. As shown in Table 2, when processing the data of the transformer area affected by common-mode high temperature attack, the cumulative contribution ratios of nodes A01, A02, and A03 all exceeded the 85% threshold, successfully participating in the principal component construction; their calculated Euclidean distance deviation feature parameters were 1.25, 2.10, and 2.85, respectively, none of which exceeded the absolute boundary set by the system statistical variance. The control logic characterized it as common-mode temperature drift caused by macro-meteorological conditions, forcibly intercepting and eliminating it, and setting the corresponding output parameters to zero. In contrast, node B01's Euclidean distance deviation feature parameter in the dimensionality reduction space reached 5.42, exceeding the feature distance deviation comparison threshold limit, exhibiting spatially isolated distribution characteristics; therefore, its association with the group climate characteristics was severed, confirming that it had undergone physical ontology degradation, and finally, the scalarized dimensionality reduction output showed a true aging bias of 3.15%.

[0098] Specific implementation instructions for step S3: Execute a closed loop of real-time automatic physical entity correction and pre-disaster prevention and maintenance scheduling:

[0099] By implementing a closed-loop physical control mechanism directly applied to the core attributes of metrology instrument calibration: Real-time metrology error parameters generated by the dimensionality reduction calculation in the preceding steps are acquired. The control unit imports these real-time metrology error parameters into a comparator, extracts their actual values ​​with positive and negative polarities, and performs a direct numerical comparison with a pre-configured legally permissible error value range. Only when the actual value of the real-time metrology error parameter is determined to be greater than the upper limit or less than the lower limit of the range, thus not falling within the legally permissible error value range, is the subsequent generation and processing logic activated; this smoothly transitions the system data flow from the calculation and evaluation layer to the control-driven stage of distributing data to the physical hardware. The legally permissible error value range is preset and read in, denoted as... This parameter characterizes the static physical upper and lower tolerance boundaries of the legally permissible fluctuations in the operating error of a smart energy meter, and it directly serves as the pre-enable switch for driving subsequent communication correction commands. In this preferred embodiment, its value is configured as a legal closed set with an absolute value limited to ±2%. If this range is set smaller, it is easier to cause global data flow oscillations, and the control loop will frequently issue invalid reverse compensation commands to devices within the normal physical drift range that are subject to minor electromagnetic interference. If the range is set larger, it is easier for actual failure error data that violates legal measurement regulations to be identified as legally allowed in the system.

[0100] The actual value of the real-time measurement error parameter is compared with the pre-configured legal error allowable range. Perform a numerical comparison operation.

[0101] When it is determined that the actual value of the real-time measurement error parameter does not fall within the legally permissible error range. Within this timeframe, it is determined that the physical hardware has experienced a physical measurement deviation, triggering a hardware physical correction intervention:

[0102] The static error offset feature value contained within the real-time measurement error parameters is extracted. This static error offset feature value is the pure percentage actual error value output after eliminating interference from the group's environmental temperature drift and undergoing standardization and dimensionality reduction in the previous steps. A measurement pulse constant compensation factor is introduced, denoted as... This refers to the digital control step size adjustment parameter necessary to reverse the bias of the internal analog-to-digital converter (ADC) sampling accuracy of a physically misaligned energy meter and pull it back to the legally expected reference zero level. In this embodiment, its value is set as an integer fine-tuning amount of 1 to 5 pulses, calculated based on the reference energy constant mapping.

[0103] Obtain the reference pulse constant of the instrument under test (denoted as ). The unit is imp / kWh. In this embodiment, the reference pulse constant is preferably the basic nameplate constant of physical meters commonly used in the low-voltage power supply and distribution field (for example, specifically configured as 1600imp / kWh or 3200imp / kWh), and it is directly converted into the metering pulse constant compensation factor using basic algebraic operations. The specific calculation steps are as follows: It extracts the percentage feature value of the calculated real-time static error offset. (For example, the current intrinsic aging deviation of the meter causes the actual deviation percentage to be -0.25%); multiply this characteristic value by negative one to extract its opposite, thereby achieving reverse value cancellation; and use the reference pulse constant. Divide the value by the constant 100 to obtain the quotient; multiply the extracted opposite number by the quotient; perform a rounding operation to remove decimal places, mathematically expressed as ROUND(), which directly removes and discards all decimal places from the result of the above multiplication, and outputs an integer control step size as a compensation factor for the metering pulse constant. .

[0104] Metering pulse constant compensation factor The data is serialized and encapsulated into the data payload segment of a dedicated downlink low-power wireless or broadband carrier communication control message to generate an energy meter error calibration command with action guidance attributes.

[0105] The error calibration command for the energy meter is sent via the downlink communication physical interface channel of the edge IoT gateway; the electrical signal of the energy meter error calibration command drives the internal metering processing unit of the energy meter under test to execute a hardware-level response action: controlling the processing unit to load the metering pulse constant compensation factor. This process overwrites the original configuration register of the internal metrological pulse constant for the current physical measurement cycle. The physical action of updating the register value completes the hard-core technical flow from pure algorithmic state derivation to substantial correction and calibration of the physical electronic device in a closed-loop, online manner without power interruption. While executing the pulse calibration sequence for a single micro-measurement cycle, a predictive disaster prevention monitoring logic closed loop is triggered in parallel:

[0106] Going back along the historical timeline, extract numerical snapshots of multiple real-time measurement error parameters that were calculated and locally stored by preset timed tasks within several consecutive historical testing cycles (the past 30 consecutive diagnostic days).

[0107] Introduce a set of historical error records, denoted as This multidimensional and complex dataset constitutes a sequence of slices representing the historical error states of individual electricity meters, arranged continuously over time. It maps the irreversible physical aging trajectory of metering assets over a relatively long period due to accumulated electrothermal stress. Within the current macro-level scheduling network architecture, this set of historical error records... The data is synchronously pushed to the remote master station via the wide area network uplink private network channel and serves as a long-term basis feature source flow for the predictive algorithm analysis engine deployed in the upstream operation and maintenance scheduling server (this engine is a fault trend prediction and status assessment software module running in the operation and maintenance scheduling server processor; its specific mathematical analysis and one-dimensional fitting derivation rules are fully defined by the least squares extreme value constraint equations described later). This is used by the predictive algorithm analysis engine to perform long-term component lifespan degradation trend assessment. Based on the extracted snapshots of multiple real-time measurement error parameters, a horizontal aggregation and splicing operation is performed along the time axis to construct a historical error record set. .

[0108] Historical error record set The time span slices within the timeframe are used to perform feature fitting calculations. Specifically, the discrete timestamps of each historical detection period are extracted as a sequence of independent variables, and the corresponding real-time measurement error parameter snapshot values ​​are extracted as a sequence of dependent variables. For these two sets of vector sequences, the least squares method is used to construct an optimal one-dimensional linear regression model. This embodiment quantitatively evaluates the overall fit between the regression line and the actual discrete error data points, establishing a residual sum of squares extreme value constraint objective function, denoted as... Its algebraic expression is as follows:

[0109]

[0110] The control logic applies the slope parameter in the objective function. Calculate the partial derivatives of the bias intercept parameter b and the bias intercept parameter b respectively. and The partial derivatives of the two equations are forced to be zero to locate the extreme point where the sum of squared residuals is minimized. The offset intercept parameter *b* physically represents the initial inherent fundamental error of the energy meter under test at the start of the current statistical cycle. By simultaneously solving the two partial derivative equations algebraically, the offset intercept parameter *b* is used as a global boundary constraint and directly eliminated and integrated into the independent analytical expression of the slope parameter. The simplified slope analytical solution calculation rules, based on the time and error discrete observation array and performing arithmetic operations, are then directly sent to the calculation module.

[0111]

[0112] Substituting the entire stored discrete sequence data into the analytical solution calculation rules described above, the values ​​are directly multiplied and summed to extract the slope value corresponding to the global extremum point that minimizes the objective function. This slope is then used as the initial geometric derivative component of the regression line. The discrete parameters... The meaning is the specific timestamp snapshot corresponding to the i-th diagnostic task. Before performing the derivative of the objective function, a relative time standardization operator is introduced: This involves reading the baseline timestamp snapshot of the first diagnostic task, subtracting this baseline timestamp from each subsequent specific timestamp snapshot, and dividing the resulting absolute difference by a fixed constant time unit per day (preferably 86400 seconds in this embodiment), thereby standardizing the discrete parameters. The physical dimension is converted into a scalar value representing the duration of relative operational evolution in consecutive days; parameter This parameter represents the actual percentage of the real-time measurement error scalar output for this diagnostic task. and These represent the timestamp characteristics and the global arithmetic mean state characteristics of the error values ​​within the current statistical period, respectively. The constant n represents the total number of discrete snapshot points of the input operator; considering the instrument's storage cycle constraint, 30 consecutive recording points are preferred in this embodiment.

[0113] The scheduling and control node performs a time-scale standardization mapping action: Specifically, by obtaining the fixed number of days parameter of a standard natural month (configured as a fixed value of 30 days in this embodiment), the initial geometric derivative component is directly multiplied with the fixed number of days parameter to uniformly amplify and align its time dimension to the single-month level; the multiplied and amplified result is output and solidified as a parameter representing the monthly acceleration rate of physical aging and degradation of the core measurement components inside the energy meter under test, representing the slope value of error change.

[0114] The scheduling system extracts the pre-configured critical slope for safety decay, denoted as... This parameter is based on the benchmark replay of offline historical failure data streams: by injecting the accelerated aging historical log features of damaged and scrapped energy meters at the end of their lifespan into the feature library, the drift slope of each stage is dynamically compared with the early warning time window of the final burn-out mutation; ensuring coverage of sudden failures and reserving at least four weeks of buffer time are the convergence extraction conditions, and finally the geometric rate of a single month's positive absolute drift rate reaching 0.15% is extracted and solidified as the critical slope for safe degradation. .

[0115] This embodiment further demonstrates the effectiveness of long-term baseline feature source flow fitting in disaster prevention and dispatching. It extracts a snapshot of the real storage repository from the historical normal operation of the distribution area (containing discrete metering error evolution data of multiple meters over the past 30 consecutive natural days), injects these data streams, and performs linear regression fitting. The corresponding relationship between the monthly acceleration rate of physical aging degradation and dispatching actions is shown in the table below:

[0116] Table 3: Fitting and Scheduling Output Results of the Aging Evolution Slope of Energy Meters under Historical Snapshots of the Runtime Library

[0117] Physical tags for high-risk energy meters under test Enter the historical record slice span (calendar days). Initial fit geometric derivative components (daily deviation rate) Error change slope (unitary monthly acceleration rate) Safety decay critical slope Disaster prevention dispatch and dispatch control actions Meter C-008 30 0.00% 0.03% 0.15% Silent security monitoring (not triggered) Meter C-012 30 0.00% 0.09% 0.15% Silent security monitoring (not triggered) Meter C-055 30 0.01% 0.18% 0.15% Forced triggering of work order issuance Meter C-089 30 0.01% 0.36% 0.15% Forced triggering of work order issuance

[0118] The disaster prevention dispatching and control action is to extract the calculated uniform monthly error change slope value and compare it with the pre-configured safety degradation critical slope value. When the slope does not exceed the limit, a "safety silent monitoring" Boolean value is output to keep the state machine in sleep mode; when the slope exceeds the limit, a "force trigger work order issuance" Boolean value action instruction is output.

[0119] Table 3 shows the input of 30 consecutive data points based on historical snapshots of the repository. Strict extreme value constraint equations were constructed using the least squares method, and the initial daily geometric derivative components were extracted. For meters C-008 and C-012, the error change slopes after their dimensions were uniformly amplified to the monthly level were 0.03% and 0.09%, respectively, both within the safety degradation critical slope of 0.15%. This indicates they conform to the normal natural physical wear and tear lifecycle of components and do not generate action commands, thus preventing invalid dispatch orders from overwhelming network resources. However, the slope fitting values ​​for meters C-055 and C-089 reached 0.18% and an extremely steep 0.36%, respectively, exceeding the safety degradation critical slope constraint threshold. It was directly determined that the resistive elements inside these two physical devices had experienced irreversible thermal stress breakdown, thus triggering a status diagnosis association tag and forcibly generating an upstream investigation work order.

[0120] The calculated slope of error change is compared with the pre-configured critical slope for safe degradation. Perform a numerical magnitude and directional condition comparison and verification. When the slope of the judgment error change exceeds the critical slope of safety degradation... At that time, it is determined that the physical components have entered a period of accelerated degradation and deterioration, triggering a pre-disaster prevention physical maintenance scheduling action: generating on-site inspection and repair instructions indicating the rapid collapse trend of aging and degradation of hardware components. The on-site inspection and repair instructions are formatted, packaged, and then synchronously pushed upstream as a status diagnosis association tag on the target output side. The control status diagnosis association tag flows upward along the wide area network's uplink private network channel to directly trigger the information layer's maintenance scheduling server to automatically parse the association tag and drive the upstream dispatch information system to automatically generate a proactive inspection physical entity work order for high-risk energy meters under test.

[0121] In this embodiment, the time-varying boundary relaxation correction factor The output range is defined between [1, 1.5].

[0122] When the time-varying boundary relaxation correction factor approaches 1, it indicates the current extracted out-of-bounds duration statistic. The preset hardware thermal overload tolerance time constant has not been exceeded. This status indicates that the current distribution area is experiencing a short-term local environmental change or a sudden overcurrent overheating of a single meter. Under this status, the steady-state cutoff reference threshold is maintained to prevent potential physical hardware burnout with maximum sensitivity.

[0123] When the time-varying boundary relaxation correction factor approaches 1.5: this represents the out-of-bounds duration statistic. The system exhibits long-term dwell characteristics, indicating an abnormal weather pattern (such as a prolonged heat wave). At this point, the temperature drift of the physical electronic components has become a common-mode condition across the entire distribution area. Time-varying boundary relaxation correction factor. Approaching 1.5 indicates that the system is proactively implementing maximum physical compliance concessions, and by significantly expanding the tolerance boundary, it ensures that network-wide paralyzing false alarms will not be triggered due to normal weather conditions.

[0124] Metering pulse constant compensation factor The output range is constrained to discrete integers in the range [1, 5]. The metering pulse constant compensation factor approaches 1, indicating that the hardware has just exceeded the legally permissible error range. The boundary is in an early physical degradation state with slight inaccuracies. The metering pulse constant compensation factor approaches 5, indicating that the analog-to-digital converter (ADC) of the metering device has experienced physical drift. The physical hardware is determined to have reached an irreversible damage state and is taken over by the disaster relief dispatch single link. There is a nonlinear logarithmic positive correlation between the over-limit duration statistic and the time-varying boundary relaxation correction factor, including the activation dead zone. The temperature drift of semiconductor components subjected to prolonged exposure to high ambient temperatures exhibits thermodynamic characteristics of initial nonlinear divergence followed by a tendency towards a saturation stability limit. The natural logarithm operator is introduced to map this physical phenomenon; setting a time constant dead zone of 72 hours logically isolates "brief overcurrent" from "climate dwell".

[0125] This embodiment is configured in the following meteorological station digital twin monitoring scenario: a set of metering nodes for a virtual distribution area is set up, and a fixed hardware thermal overload tolerance time constant is defined for the system. The smooth step time constant is 72 hours. For 24 hours, the compensation amplification factor The value is set to 0.15. A continuous stream of high-temperature meteorological data is injected into the edge IoT gateway processor. The control module extracts the over-limit duration statistics in real time and calls the tolerance boundary dynamic expansion control logic to execute feature flow and calculation updates. After the edge processor determines that the over-limit duration statistics exceed the hardware thermal overload tolerance time constant, it extracts the time difference of the excess portion, divides it by the smoothing step time constant to construct a logarithmic extrapolation base; then, after natural logarithmic transformation and multiplication by the compensation amplification factor, it outputs the time-varying boundary relaxation correction factor; finally, it performs scalar multiplication with the statically configured steady-state truncation benchmark threshold, issues and overwrites the current compensation tolerance boundary value. This embodiment illustrates the evolution law of internal state variables in response to long-term meteorological residence, extracting the following discrete working condition slices for theoretical calculation and parameter mapping extrapolation: Table 4 aims to show the adaptive widening trajectory of the core tolerance boundary interception network in the physical process of gradually evolving from a normal state to long-term extreme climate residence. By inputting different over-limit duration statistics parameters, the output time-varying boundary relaxation correction factor and the final legal boundary are determined.

[0126] Table 4: Example of Calculation of Tolerance Boundary Compensation Index under Macro-meteorological Stationing Conditions

[0127] Macro-meteorological stationing sequence Duration of exceeding the limit (in hours) Dynamic differential time quotient Time-varying boundary relaxation correction factor Steady-state cutoff benchmark threshold (%) Compensation tolerance boundary value (%) Improvement in false positive interception rate (%) Operating Condition 1: Conventional Dwelling Test Point 24 Inactive dead zone 1 8.00% 8.00% 0.00% Operating Condition 2: Extreme Value Locking Critical Point 72 0 1 8.00% 8.00% 0.00% Operating Condition 3: Initial Point of Logarithmic Expansion 96 1 1.1 8.00% 8.80% 10.00% Condition 4: Harsh Deployment Point 144 3 1.21 8.00% 9.68% 21.00% Operating Condition 5: Extreme Deepwater Dwelling 240 7 1.31 8.00% 10.48% 31.00% Operating Condition 6: Ultra-long Stay Convergence Point 312 10 1.36 8.00% 10.88% 36.00%

[0128] The improvement in false positive interception rate is achieved by extracting the absolute value of the calculated compensation tolerance boundary value, subtracting the absolute value of the preset steady-state cutoff benchmark threshold, and obtaining the absolute difference between the two. This absolute difference is then divided by the absolute value of the steady-state cutoff benchmark threshold to obtain the improvement quotient. This quotient is multiplied by a constant of 100, and a standard percentage sign is appended to the value, thus mapping it from an absolute difference dimension to a one-dimensional quantitative characteristic indicator with pure percentage dimensions. This parameter quantification, when meteorological conditions cause common-mode physical temperature drift across the entire network, can dynamically widen the legal monitoring boundaries to avoid the relative spatial expansion ratio that prevents normal equipment from being mistakenly judged as physically failed.

[0129] Comparing the output responses of Condition 2 (72 hours) and Condition 4 (144 hours), it can be seen that when the duration of the out-of-limit operation does not exceed the hardware thermal overload tolerance time constant, the 8% interception band is maintained; however, when the climate anomaly exceeds 72 hours, the internal state machine triggers a smooth switch. In existing conventional technical solutions, facing the scorching heat of Condition 4 (144 hours), a rigid 8% verification is still adopted. This would result in all metering devices that generate a normal temperature drift greater than 8% due to thermodynamic laws being fully intercepted, causing false alarms and paralysis. In the solution of this invention, by executing feature dimensionality reduction logic containing the natural logarithm operator, the boundary parameters have been adaptively extrapolated to 9.68% in Condition 4. Based on the aforementioned derivation logic calculation, it can be seen that compared with the existing static technical benchmark value of 8%, the false positive interception rate has been improved by 21%. Further comparison of the data trajectories of Condition 5 and Condition 6 shows that as the time continues to worsen from 240 hours to 312 hours, the compensation tolerance boundary value slowly converges from 10.48% to 10.88%. This data trajectory verifies that the system boundary does not expand linearly over time.

[0130] Figure 5 The X-axis represents the duration of exceeding the limit, the Y-axis on the left represents the compensation tolerance boundary value, and the Y-axis on the right represents the improvement in the false positive interception rate. Here, T1 represents the critical value of the time constant (72); V0 represents the steady-state cutoff baseline threshold value (8); V1 represents the absolute limit boundary value of the model's asymptotic convergence state (10.88); and P2 indicates the maximum interception improvement limit value of 36 at the end of the validation sequence. Figure 5 Numerical calculation results show that when the parameter crosses the dead time constant boundary T1 of 72, the tolerance boundary changes from a hard-locked state to an adaptive state of nonlinear asymptotic smooth expansion constrained by the natural logarithm operator, and finally converges to the limit point V1.

[0131] Figure 1The bottom entity perception layer displays "1D temperature data" graphic labels showing the corresponding ambient temperature feature data of multiple energy meters in the overall physical scene of the metering box. Physically, it not only corresponds to and supports the acquisition of multi-channel voltage and current sampling sequences in step S1, but also corresponds to the action logic chain of each step component in the top process box sequence in the edge gateway processing layer in terms of causal logic: the feature processing side component logic of step S1, heterogeneous perception and spatiotemporal feature alignment mapping; the baseline construction component and tolerance correction mapping logic of the first half of step S2, group baseline construction and dynamic tolerance expansion mapping; the feature calculation side component logic of principal component dimensionality reduction decoupling and error calculation mapping; the sequence of energy difference values ​​exceeding the tolerance range after the energy meter under test exceeds the boundary is projected to the principal component feature vector set extracted after orthogonal decomposition for spatial dimensionality reduction, and the conversion action of multidimensional vector to one-dimensional scalar is performed based on spatial geometric modulus calculation and full-scale ratio mapping; finally, the real-time metering error parameter representing its actual measurement deviation is calculated and output; the hardware physical correction and operation and maintenance scheduling closed-loop mapping step S3 closed-loop output side component logic. The "Status Diagnosis Tag" points upwards to the top macro-level scheduling layer operation and maintenance server; it also visualizes the generated electricity meter error calibration instructions and maps them downwards to the electricity meter.

[0132] Figure 2 This demonstration shows how data is acquired from multiple electricity meters within a low-voltage power distribution metering box—a "metering box acquisition node." It also shows the physical characteristics of the raw voltage and current waveforms exhibiting misalignment on the horizontal time axis due to narrowband communication network latency. The intermediate edge IoT gateway module performs asynchronous sampling time difference analysis based on preset node power conservation rules. A dynamic time window matching processing component, represented by a bar chart, performs timestamp benchmark alignment on the multi-channel voltage and current sampling sequences at the physical level, eliminating waveform misalignment errors caused by transmission delays in the physical network. The gateway outputs a scaled-down "time-synchronized electricity characteristic sequence" that can be directly compared within the same physical time dimension.

[0133] Figure 3A "multi-node error covariance matrix" characterizing the comprehensive error distribution of a group of devices within the same power supply topology is constructed, and orthogonal decomposition is performed on it. The three-dimensional coordinate system in the figure maps the feature space of the constructed covariance matrix; the "principal component feature vector set (first component)" and "principal component feature vector set (second component)" labeled on the X and Y axes respectively, concretely demonstrate the set of basis vectors extracted after the orthogonal decomposition operation, which are mutually orthogonal and represent different error evolution directions. The "electricity difference sequence exceeding the tolerance range" corresponding to the energy meter under test is obtained and projected onto the aforementioned principal component feature vector set. The densely distributed point cloud in the center of the image represents the feature component whose Euclidean distance deviation feature parameter calculated after projection is less than or equal to the corresponding distance threshold. This component is identified as the "common mode temperature drift error component caused by external meteorological temperature changes" and filtered out. The isolated outward-extending arrow maps the "spatial projection of the electricity difference sequence exceeding the tolerance range" where the Euclidean distance deviation feature parameter exceeds the absolute distance boundary. The "geometric modulus" of this vector is extracted and converted into a scalar to calculate the true metering error parameter excluding environmental interference.

[0134] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for online verification and diagnosis of meter box energy meters based on multi-source measurement data fusion, characterized in that, The specific steps include: S1: Obtain the multi-channel voltage and current sampling sequence and the corresponding ambient temperature feature data reported by the metering box acquisition node, and perform timestamp reference alignment and reference ratio conversion on the multi-channel voltage and current sampling sequence to output the time-synchronized power feature sequence. S2: Perform an arithmetic mean calculation on all time-synchronized power characteristic sequences under the same topological association to extract the population power fluctuation mean curve, and perform tolerance mapping correction on the population power fluctuation mean curve based on the ambient temperature characteristic data to construct a temperature-compensated population reference baseline. Based on the environmental temperature characteristic data, the average curve of the population's power fluctuation is corrected by tolerance mapping to construct a temperature-compensated population reference baseline, including: Extract the historical average load rate characteristic data of the energy meter under test; In response to the determination that the historical average load rate characteristic data exceeds the preset rated high load heating threshold, a heat penalty attenuation weight is assigned to the energy meter under test based on the ambient temperature characteristic data to block the benchmark calculation deviation caused by the local self-heating of the energy meter under test. By combining the preset equivalent coefficient of temperature rise of meter components, the ambient temperature characteristic data is mapped to a global temperature compensation weight value. The global temperature compensation weight value and the heat penalty attenuation weight are multiplied and weighted. The weighted result is used to perform dynamic proportional stretching calculation on the upper and lower limit fluctuation tolerance range of the average fluctuation curve of the group power fluctuation. In response to the determination that the stretched tolerance boundary exceeds the hardware operation safety extreme value threshold, the tolerance cut-off protection logic is triggered to forcibly solidify the current boundary, thereby generating the temperature-compensated group reference baseline containing the dynamic allowable tolerance range. The time-synchronized energy characteristic sequence of the energy meter under test is compared with the time-series curve value of the temperature-compensated group reference baseline to calculate the real-time metering error parameter characterizing its actual measurement deviation. S3: Perform a size comparison between the actual value of the real-time metering error parameter and the pre-configured legal error value allowable range. In response to the determination that the actual value does not fall within the legal error value allowable range, generate an energy meter error calibration command for the energy meter under test based on the real-time metering error parameter. The energy meter error calibration command is configured to trigger the internal metering processing unit of the energy meter under test to perform a numerical update of the metering pulse constant and simultaneously output the corresponding status diagnosis association tag.

2. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 1, characterized in that: The acquisition of the multi-channel voltage and current sampling sequences reported by the metering box acquisition node and the corresponding ambient temperature characteristic data includes: Acquire discrete meteorological temperature sampling sequences transmitted from external meteorological interface nodes; By combining the thermal conduction delay parameter of the metering box, a first-order low-pass filtering algorithm is used to perform smooth interpolation processing on the discrete meteorological temperature sampling sequence to generate the ambient temperature feature data that matches the multi-channel voltage and current sampling sequence in terms of time granularity.

3. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 2, characterized in that: Perform timestamp reference alignment and reference scaling conversion on the multi-channel voltage and current sampling sequences to output a time-synchronized power characteristic sequence, including: Based on the preset node power conservation rule, the asynchronous sampling time difference in the multi-channel voltage and current sampling sequence is analyzed, and dynamic time window matching processing is performed to remove the waveform misalignment error caused by the transmission delay due to network transmission time lag. After removing the waveform misalignment error caused by the transmission delay, the measurement data stream is scaled using a voltage and current unified dimension conversion ratio to extract the time-synchronized power feature sequence that can be directly compared in the same time dimension.

4. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 3, characterized in that: The step of performing a time-synchronized energy characteristic sequence comparison operation between the energy meter under test and the temperature-compensated group reference baseline to calculate the real-time metering error parameter characterizing its actual measurement deviation includes: The time-synchronized energy characteristic sequence of the energy meter under test is projected onto the temperature-compensated group reference baseline containing the dynamic allowable tolerance range for time-series curve numerical comparison. In response to determining that the time-synchronized power characteristic sequence of the energy meter under test exceeds the boundary of the dynamic allowable tolerance range, the power difference value exceeding the tolerance range in the part exceeding the corresponding boundary is extracted. The power difference exceeding the tolerance range is subjected to high-frequency electrical noise smoothing filtering to remove common-mode interference components caused by grid voltage transients. Principal component extraction is used to reduce the dimensionality of the real-time metering error parameter characterizing the physical aging drift of the metering element of the energy meter under test.

5. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 4, characterized in that: Based on the real-time metering error parameters, an energy meter error calibration command is generated for the energy meter under test, including: The actual value of the real-time measurement error parameter is compared with the pre-configured legal error value allowable range. In response to the determination that the actual value of the real-time measurement error parameter does not fall within the legally allowed error value range, the static error offset of the real-time measurement error parameter is extracted, and the measurement pulse constant compensation factor is extracted based on the multiplication of the static error offset with the reference pulse constant of the current instrument under test and the rounding operation of truncating the decimal places. The metering pulse constant compensation factor is encapsulated in a downlink communication control message to generate the energy meter error calibration command, thereby driving the internal metering processing unit of the energy meter under test to load the metering pulse constant compensation factor and overwrite the metering pulse constant of the current measurement cycle.

6. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 5, characterized in that: The status diagnosis association tags corresponding to the synchronized output include: Extract multiple real-time measurement error parameters calculated within multiple consecutive historical detection cycles to construct a historical error record set; Perform error change trend slope fitting calculation on the historical error record set to generate error change slope value characterizing the physical aging evolution rate of the internal measuring components of the energy meter under test; In response to the determination that the slope value of the error change exceeds the preset safety degradation threshold slope, an on-site inspection and maintenance instruction characterizing the aging and degradation trend of the component is generated. The on-site inspection and maintenance instructions are output as the status diagnosis association tags, which are configured to trigger the operation and maintenance scheduling server to generate inspection instructions for the energy meter under test.

7. The online verification and diagnosis method for meter box energy meters based on multi-source measurement data fusion according to claim 6, characterized in that: The trigger tolerance truncation protection logic, which forcibly solidifies the current boundary, includes: For the state where the dynamic allowable tolerance boundary after stretching exceeds the hardware operation safety extreme value threshold, a time accumulation statistical operation is performed to extract the over-limit duration statistic used to characterize the degree of high temperature over-limit; The above-limit duration statistic is compared with the pre-configured hardware thermal overload tolerance time constant. In response to determining that the over-limit duration statistic is less than or equal to the hardware thermal overload tolerance time constant, the static extreme value locking execution logic is activated; the locally configured steady-state cutoff benchmark threshold is retrieved as the absolute constraint extreme value, and the absolute constraint extreme value is used to perform a constant assignment operation on the dynamic allowable tolerance boundary, locking it as a fixed extreme value constant; In response to the determination that the over-limit duration statistic is greater than the hardware thermal overload tolerance time constant, the tolerance boundary dynamic expansion control logic is triggered; based on the division of the over-limit duration statistic with the preset step time constant and the natural logarithm extraction operation, the time-varying boundary relaxation correction factor is calculated. The time-varying boundary relaxation correction factor and the steady-state truncation benchmark threshold are calculated by numerical scaling multiplication to dynamically generate the relaxed compensation tolerance boundary value. The relaxed compensation tolerance boundary value is used to replace and update the stretched dynamic allowable tolerance boundary, so that the updated boundary can be used as the current configuration boundary to participate in the time series curve numerical comparison calculation in subsequent periods.

8. The online verification and diagnosis method for metering box energy meters based on multi-source measurement data fusion according to claim 7, characterized in that: The real-time metering error parameters characterizing the physical aging drift of the metering element of the energy meter under test are obtained by dimensionality reduction through principal component extraction, including: Obtain the power difference sequence that exceeds the tolerance range for all associated nodes under the same topology association, perform feature data array splicing operation on multiple power difference sequences, and construct a multi-node error covariance matrix that characterizes the comprehensive error distribution characteristics of the equipment group within the same power supply topology; Perform orthogonal decomposition calculation on the multi-node error covariance matrix to extract the principal component feature vector set representing different error evolution directions; Obtain the sequence of power difference values ​​exceeding the tolerance range corresponding to the power meter under test, project it onto the principal component feature vector set, and calculate the Euclidean distance deviation feature parameter of the sequence relative to the mean center of error fluctuation of the equipment group in the distribution area after projection; The principal component feature vector set is subjected to a data separation and judgment operation using a preset feature distance deviation comparison threshold. In response to determining that the Euclidean distance deviation feature parameter is less than or equal to the corresponding distance threshold, the feature component is identified as a common-mode temperature drift error component caused by external meteorological temperature change, and is filtered out from the subsequent individual error evaluation calculation. In response to the determination that the Euclidean distance deviation feature parameter is greater than the corresponding distance threshold, feature components that present an isolated distribution state are filtered out and identified as the intrinsic aging deviation vectors of individual components that have been excluded from the interference of group environmental temperature drift. The intrinsic aging deviation vector of the individual component is extracted as the target physical quantization error feature parameter. The absolute value of the geometric modulus of the feature vector in space is calculated, and the absolute value of the geometric modulus is converted into a one-dimensional scalar value based on a preset full-scale reference, and output as the real-time measurement error parameter.

Images