Adaptive estimation method of meter status supporting data incompleteness

Abstract

The present invention relates to a meter state adaptive estimation method supporting data incomplete completion. The method collects continuous measurement data of the meter under a unified time base to form an original measurement data sequence. Based on the original measurement data sequence, a state parameter set for describing the current working state is generated. The original measurement data sequence is divided into multiple data windows, and the expected measurement value in each window is predicted based on the change amplitude and trend direction of the data collected in adjacent time periods. The predicted measurement value is compared with the actual measurement value to identify the position where the difference exceeds a set threshold and form an incomplete position index set. For the incomplete position index set, the completion value is searched within a predetermined reasonable numerical range in combination with the change trend and continuity of adjacent data to generate a completed measurement data sequence. The present invention improves the integrity and accuracy of meter data in the presence of loss or abnormality.

Description

Technical Field

[0001] The present invention relates to the technical field of electric meters, and in particular to a meter state adaptive estimation method supporting data incompleteness completion. Background Art

[0002] Existing meter data collection technologies typically collect and measure energy parameters in real time and record them as a data sequence for subsequent metering, settlement, and monitoring analysis. To improve data accuracy, some systems combine calibration parameters with historical operating data for error correction. Data caching and multi-channel redundancy mechanisms are also employed to mitigate temporary data loss.

[0003] However, in actual operation, meters operating continuously for long periods of time may still experience data loss, abnormal offsets, or incomplete segments due to communication interruptions, momentary power outages, signal interference, or internal state fluctuations. Existing technologies often rely on simple interpolation, smoothing, or single-rule correction methods for compensation. These methods fail to fully identify and utilize the dynamic operating characteristics of the meter and the correlation between previous and subsequent data. This leads to significant deficiencies in both the accuracy of data completion and state assessment, particularly impacting overall data integrity and reliability in critical metering scenarios.

[0004] Therefore, it is urgent to propose a meter state estimation and completion technology solution that can achieve higher adaptability and accuracy under data incomplete conditions. Summary of the Invention

[0005] The present application provides a meter status adaptive estimation method that supports data incompleteness and completion, so as to improve the integrity and accuracy of meter data in the presence of loss or anomalies.

[0006] The present application provides a method for adaptively estimating meter status by supporting data incompleteness completion, comprising:

[0007] Under a unified time base, the continuous measurement data of the meter is collected to form the original measurement data sequence;

[0008] Based on the original measurement data sequence, combined with calibration parameters during the meter manufacturing process and pre-collected historical stable operation data, a state parameter set is generated, wherein the state parameter set is used to describe the normal output range and change trend of the meter under the current working environment;

[0009] Dividing the original measurement data sequence into multiple data windows in chronological order, and within each data window, calculating the change amplitude and trend direction of the collected data in adjacent time periods in combination with the trend direction and change rate boundary recorded in the state parameter set, and predicting the expected measurement value in the current window;

[0010] Comparing the predicted measurement value with the actual collected measurement value, and judging, based on the reference output range and tolerance interval set in the state parameter set, if the difference between the two exceeds a set threshold, recording the corresponding position as incomplete data to form an incomplete position index set;

[0011] For the incomplete position index set, by analyzing the numerical change trend and continuity relationship of the data collected before and after the incomplete position, and searching for the supplementary value within the reasonable measurement interval defined in the state parameter set, the supplemented measurement data sequence is estimated and generated;

[0012] During the completion process, the state parameter set is corrected and adaptively updated according to the offset between the completed value and the state parameter set to output the final completed measurement data sequence, the updated state parameter set and the incomplete confidence score.

[0013] The beneficial effects of the technical solution provided by this application include:

[0014] (1) By combining data change trend analysis with continuity relationships, the present invention limits a reasonable search interval during the incomplete data completion process, effectively avoiding the error accumulation caused by traditional interpolation methods under mutation points or abnormal fluctuations, and improving the reliability and accuracy of the completed data. (2) By synchronously analyzing the offset between the completed value and the current state parameter during the completion process, the state parameter set is corrected in real time, so that the state estimation process has continuous self-adaptation capabilities, can dynamically adapt to the actual changes in the meter's operating state, and improve long-term operating stability. (3) The incomplete identification mechanism based on short-term trend prediction and threshold judgment can accurately identify the time point when the data anomaly occurs, avoid missed judgments or misjudgments, effectively improve the accuracy of the incomplete position index set, and lay a high-quality basic data support for subsequent completion processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flowchart of a meter state adaptive estimation method supporting data incomplete completion provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0016] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0017] The first embodiment of the present application provides a method for adaptively estimating meter status that supports data incompleteness and completion. Figure 1 , which is a schematic diagram of the first embodiment of this application. Figure 1A first embodiment of the present application provides a method for adaptively estimating meter status that supports data completion, which is described in detail.

[0018] Step S101: Under a unified time reference, continuous measurement data of meters are collected to form an original measurement data sequence.

[0019] In step S101, the system continuously collects data from the measurement channels of an energy meter (hereafter referred to as the meter) based on a unified time base, generating a sequence of raw measurement data covering the entire measurement cycle. This step ensures temporal consistency and physical continuity of the underlying data for all subsequent completion and estimation operations, avoiding systematic errors introduced by misaligned sampling times or inconsistent data sampling intervals.

[0020] During implementation, the meter must first be initialized and configured to ensure synchronization between its internal clock and the system master clock. This system master clock can be a substation-level time server, a central control center clock signal, or a standard time reference provided by a GPS timing module. Network timing protocols (such as NTP) or hardware pulse synchronization mechanisms ensure that the clocks of all sampling nodes are consistent within an acceptable error range. Based on this unified clock, data collection operations are triggered at fixed intervals, such as every 100 milliseconds, every second, or every 15 minutes, depending on the accuracy requirements of the actual power monitoring application. Each data collection operation captures the measured values ​​of key power parameters such as voltage, current, active power, reactive power, power factor, and frequency. Where necessary, this also includes auxiliary status information such as the meter's internal temperature, battery voltage, or communication status.

[0021] The raw measurement data should be recorded in a timestamp format to ensure that each piece of data has a clear time identifier at the sampling moment, and the sampled data should be sent to the central processing system through the local cache or uplink communication module. If there is a network interruption or data transmission anomaly during the sampling period, the raw data can be temporarily stored by the cache mechanism and uploaded uniformly after the network is restored to ensure the integrity and continuity of the data. All collected data ultimately constitute a raw measurement data sequence, which is arranged in chronological order and must not overlap or miss. The data format should include at least one time field and one or more measurement value fields to support subsequent window division, trend analysis, and completion calculation operations.

[0022] In terms of data quality control, to ensure sufficient credibility of raw measurement data, a data anomaly screening mechanism should be implemented in the acquisition module. For example, this can eliminate obviously erroneous data with zero values, identify abnormal conditions such as voltage drops and power surges, and retain these identification results in data annotation. Some meters have automatic annotation capabilities, adding a "quality tag" field to the sampled data to indicate whether the data is normal, whether it is a missing value, or whether it is a cache retransmission, thereby improving the efficiency of data judgment in subsequent processing.

[0023] In the above manner, step S101 not only completes the preliminary collection of measurement data, but also constructs a high-quality raw data foundation that can be directly used for subsequent state estimation and incomplete completion operations through measures such as unified time base, fixed period, structured format, continuity verification and abnormality preprocessing, thereby ensuring that the input data of the entire adaptive estimation process has stability, continuity and technical reliability.

[0024] Furthermore, the collecting of the continuous measurement data of the meter under the unified time reference to form the original measurement data sequence includes:

[0025] A unified time synchronization device is embedded in each meter. The device maintains local time based on a high-stability temperature-compensated crystal oscillator and aligns clocks via a time beacon periodically broadcast by an external master control node, ensuring global consistency of data collected from different meters.

[0026] After all meters are synchronized, continuous measurement data is collected in parallel from multiple channels including voltage, current, active power, reactive power, power factor, and frequency based on the set minimum sampling period. A corresponding timestamp tag is added to each channel to generate the initial measurement data fragment.

[0027] The initial measurement data fragments generated by each meter are transmitted to the edge node. Based on the local cache mechanism and timestamp aggregation algorithm, the edge node performs window alignment and format standardization on the different channel data from multiple meters according to a unified time base, forming a structured unified measurement data set.

[0028] Marking abnormal values, missing values, or mutation values ​​that cross boundaries in the unified measurement data set using data quality screening rules, calculating confidence scores based on data distribution density within a window, and dividing data segments into high-confidence segments and low-confidence segments;

[0029] High-confidence segments are stored as the backbone of the original measurement data sequence, and low-confidence segments are marked as key monitoring areas in the subsequent state modeling and incomplete recognition process. At the same time, corresponding confidence mask vectors are generated as important references in the subsequent prediction and completion links.

[0030] During the implementation of the present invention, in order to achieve accurate identification and incomplete completion of the meter status, it is first necessary to construct an original measurement data sequence with reliable time synchronization, high sampling accuracy and unified structural standards. The implementation basis of this step is that the measurement time of each meter must have global uniformity, thereby ensuring the accurate correspondence of subsequent processing logic such as window division, state modeling, incomplete identification and data completion in the time dimension. To this end, a high-stability time synchronization device is pre-embedded in each meter in the present invention. The device is based on a high-stability temperature-compensated crystal oscillator and is used to provide an accurate local clock reference. Taking into account the possible temperature fluctuations, electromagnetic interference and other factors in the working environment, the oscillator is designed to have excellent resistance to temperature drift and low phase noise characteristics, thereby ensuring the stability of the time base under long-term operation.

[0031] To achieve clock consistency across multiple meters, the present invention also introduces an external time broadcast mechanism based on a master control node. Specifically, the system employs a master control node as the global time coordination unit, which periodically broadcasts a synchronization beacon signal with the current timestamp. Upon receiving this signal, the metering device automatically performs clock alignment by calculating the relative offset between its local clock and the broadcast time, thereby achieving a unified time reference across multiple measurement points. This mechanism effectively avoids time drift caused by slight variations in the device's internal crystal oscillator frequency, providing a stable and reliable time foundation for subsequent data aggregation and state modeling.

[0032] After time alignment, each meter begins multi-channel data acquisition according to the set minimum sampling period. The data collected covers multiple key power parameter channels, including voltage, current, active power, reactive power, power factor, and frequency. A high-precision timestamp tag is appended to each channel's data record. This tag is synchronized with a unified time base, recording the moment each measurement occurred. This ensures that data from different channels and meters can be accurately matched in subsequent processing. This data acquisition process not only maintains measurement accuracy but also ensures controllability and consistency in subsequent processing at the data structure level.

[0033] After each meter completes data collection, the generated initial measurement data fragments will be transmitted to the edge node through the local communication module or the network communication interface. The edge node has a local high-speed cache mechanism that can receive and temporarily store raw data fragments from multiple meters. During the caching process, the edge node sorts different data sources based on timestamps, and uses a unified time window alignment strategy to synchronize the data of all measurement channels in the time dimension. Specifically, the edge node classifies the data with adjacent timestamps falling into the same window interval according to the set window width, and retains or interpolates the data that crosses the window boundary to form a measurement data record with a unified structure. At the same time, through the triple indexing mechanism of channel identification, device ID and timestamp, it is ensured that each piece of data has accurate spatiotemporal positioning information, making the logical structure between the data highly clear.

[0034] After completing time alignment and format unification, the edge node performs quality screening on the generated unified measurement data set. The goal of quality screening is to identify outliers, missing values, or mutation points that may exist in the data and explicitly mark them. To this end, the system has a built-in set of data quality screening rules, including but not limited to measurement value range limitations, mutation rate boundary constraints, and physical consistency verification between channels. For example, when the difference between the current measurement value of a channel and the previous time point exceeds the set mutation threshold and lacks abnormal collaborative features from other channels, the data is marked as a cross-boundary mutation. For example, if the voltage exists but the current is zero at a certain moment, and the power factor surges abnormally, the data set may be judged as a device sensing anomaly or data interruption phenomenon. These rules can be dynamically adjusted according to the actual deployment scenario and can be adapted and optimized based on background information such as meter model and wiring method.

[0035] After completing anomaly detection and labeling, the edge node further evaluates the credibility of the data segment based on the distribution characteristics of the measured values ​​within each time window. The credibility is calculated based on multi-dimensional feature fusion, including indicators such as the value range distribution density within the local window, the standard deviation fluctuation range, the residual anomaly rate, and the synchronization between channels. For example, when multiple channels in a certain time window show high-frequency fluctuations, obvious cross-residual deviations, and concentrated distribution of instantaneous distortion, the credibility score of the window will be significantly reduced; on the contrary, if the distribution of measured values ​​in this time period is continuous and stable, the residual fluctuations are within a controllable range, and the change trends between channels are consistent, then the window can be judged as a high-confidence interval. The system will divide all data segments into two categories: high-confidence segments and low-confidence segments based on the scoring threshold.

[0036] High-confidence segments will be saved as backbone data first and enter the subsequent state modeling process to build a reliable set of state parameters. For low-confidence segments, the system will not discard them immediately, but will clearly mark them as areas to be monitored. These areas play a key role in the subsequent incomplete identification and completion process. To this end, the system will also generate a corresponding confidence mask vector for each segment. This vector marks the confidence level of each data point in chronological order and provides auxiliary labeling dimensions, such as the degree of mutation, the length of the missing interval, and other features. This mask will be called in the subsequent prediction and completion links as a decision-making basis for core operations such as trend fitting boundary adjustment, data fusion weight allocation, and completion point priority sorting.

[0037] Step S102: Based on the original measurement data sequence, combined with calibration parameters in the meter manufacturing process and pre-collected historical stable operation data, a state parameter set is generated. The state parameter set is used to describe the normal output range and change trend of the meter in the current working environment.

[0038] In step S102, the key goal of the system is to generate a set of state parameters that can dynamically characterize the current measurement behavior of the meter, which is used to guide subsequent key links such as prediction, incomplete identification, complete inference, and parameter correction. This step is not simply to record some statistics, but to integrate the manufacturing information from the meter with the data characteristics accumulated over a long period of time in the operating environment to form a dynamically adjustable and clearly structured parameter framework that can be repeatedly called and updated in subsequent processing. The generation process of the state parameter set needs to ensure its physical rationality, temporal consistency and engineering feasibility, and the structural design should match the actual use in steps S103 to S106 to ensure the technical closed loop of the entire estimation method and the complete transmission of parameter logic.

[0039] First, the system needs to load the meter's manufacturing calibration parameters. These parameters are generally provided by the equipment manufacturer and include the linear response coefficient, zero offset, temperature drift correction factor, communication response delay, frequency correction factor, and other parameters for each measurement channel (such as voltage, current, and active power) under standard conditions. These calibration parameters are typically stored in the meter firmware as structured configuration files or register data or can be read through the communication port. The system uses these parameters as initial static correction factors to standardize the subsequently collected raw data. This ensures that the data aligns with the standard measurement system in terms of dimension, response curve, and error distribution, ensuring engineering consistency of the underlying data for subsequent trend analysis and anomaly assessment.

[0040] Next, the system uses historical stable data collected from the meter's continuous operation in the current installation environment for a period of time (e.g., 7 or 30 consecutive days) to extract time series features and perform statistical modeling on this data. To prevent periodic fluctuations in electricity consumption from interfering with the model, the system uses a sliding time window approach to segment the historical data and extract measurement intervals with strong stability and consistent trends as reference samples. After analyzing these samples channel by channel, the system extracts the following core features, which are used to construct the main structure of the state parameter set:

[0041] The first category is the reference output range for each measurement channel. This range is defined as the range defined by the mean of all sampled values ​​during stable operation plus or minus two standard deviations. This range reflects the normal fluctuation range of the channel. In subsequent steps, this range will serve as a boundary constraint for prediction error assessment and verification of the validity of supplemented values. This parameter also assists in outlier rejection and out-of-limit alarms.

[0042] The second category is the trend direction identifier, which indicates whether a channel exhibits a sustained upward, downward, or stable trend during the current operating cycle. The system performs a first-order time series difference on historical data and statistically analyzes the signs of changes over multiple consecutive cycles. For example, if a channel exhibits a positive slope at most time points and the changes are stable, it is considered an "upward trend"; if the slope is negative most of the time, it is labeled "downward"; if the signs alternate but the fluctuations are small, it is labeled "stable" or "mildly oscillating." This trend direction serves as a trend reference in the measurement prediction in step S103, guiding the direction of data extension.

[0043] The third category is the rate of change bound, which specifies the maximum rate of increase and decrease of the value per unit time for each measurement channel under normal operating conditions. This parameter is calculated by calculating the distribution of unit time differences in historical data, typically taking the 95th percentile as the upper bound and the 5th percentile as the lower bound. This parameter is used to limit the magnitude of data changes during the prediction and completion process, preventing illogical judgments when the system experiences short-term fluctuations or sampling errors.

[0044] The fourth category is the measurement response hysteresis factor, which describes whether there is a time delay in the response of a measured parameter to external load changes or power supply disturbances. For example, for current signals, the response of some loads to power switching is not immediately reflected in the current waveform, resulting in a lag of one to two sampling cycles. The system identifies lag times in event sequences in historical data and constructs a lag table for each channel. This parameter provides forward or backward shift corrections during completion and prediction, enhancing the temporal consistency of the model.

[0045] The fifth category is the fluctuation tolerance level, which measures whether a channel is susceptible to interference or exhibits significant fluctuations. The system assesses the fluctuation amplitude of a channel using the ratio of the standard deviation to the mean (coefficient of variation), categorizing it as "high tolerance," "medium tolerance," and "low tolerance." The fluctuation tolerance level influences the error threshold setting for subsequent incompleteness detection. For example, for "high tolerance" channels, a larger prediction deviation is permitted; for "low tolerance" channels, the error limit should be strictly controlled to prevent normal data from being misidentified as anomalies.

[0046] In addition to the basic state parameters mentioned above, the system can also incorporate extended items into the state parameter set, such as the mean of the previous cycle's prediction residuals, the goodness-of-fit score history during the completion process, and the count of consecutive anomaly detections, to support adjustments to the adaptive judgment logic. These extended parameters do not directly participate in the prediction or completion process, but can serve as controller parameters to influence the subsequent update strategy of the state parameter set. Furthermore, the system defines a tolerance range for each channel in the state parameter set, describing the allowable short-term fluctuation range of the system during different operating stages. This tolerance range is not directly derived from historical average statistics, but is instead determined based on multiple factors, such as the distribution of deviations between predicted and actual values, the characteristics of the variation cycle, and transient transient response. For example, when a channel is in the "high fluctuation tolerance level," its tolerance range can be set to ±10%; while in the "low fluctuation tolerance level," it can be limited to ±2%. This tolerance range, along with the prediction error, forms the basis for incomplete judgment.

[0047] All state parameters are stored in a structured data format, one group per channel. Each group includes the parameter value, parameter calculation basis, weight level, and update timestamp. This set is passed to steps S103-S105 as an important basis for prediction, judgment, and valuation. In step S106, feedback correction and adaptive update are combined with the completion results to form a complete dynamic closed-loop system.

[0048] By implementing this step, the system not only establishes a quantitative understanding framework for the current operating behavior of the meter, but also builds a state guidance mechanism that is tightly coupled with the data processing process, so that each subsequent judgment and operation has a basis rather than being executed in isolation.

[0049] Furthermore, based on the original measurement data sequence, combined with calibration parameters in the meter manufacturing process and pre-collected historical stable operation data, a state parameter set is generated, including:

[0050] Extracting archived factory calibration parameter sets from the meter manufacturing process, including static error curves, multi-point calibration gain coefficients, phase correction offset values, and characteristic temperature drift models, to construct an initial performance model of the meter. The initial performance model describes the basic response characteristics of the measurement output under ideal operating conditions.

[0051] Mapping and matching the initial performance model with the operating environment conditions of the meter's current deployment location, obtaining environmental characteristic variables including annual average temperature, humidity, grid voltage fluctuation range, and typical load curve by calling the environmental database of the deployment area, constructing an environmental adaptation model, and correcting drift terms in the initial performance model that may be caused by environmental changes;

[0052] The collected raw measurement data series are grouped according to different data channels. Based on the fluctuation range, change trend and frequency domain response characteristics of each channel during the historical stable operation period, statistical stability indicators and change rate indicators are extracted to construct a dynamic operation feature set for actual operating conditions.

[0053] Based on the dynamic operation feature set, an interval overlapping optimization algorithm is used to jointly calibrate the initial performance model and the fluctuation characteristics in the current operation data to generate a reference output range interval covering the time series, trend direction boundary, change amplitude upper limit, tolerance correction factor and reasonable measurement interval definition, thereby forming a structured state parameter set;

[0054] The generated state parameter set is encapsulated in a standardized manner, and a credibility identifier and update timestamp are attached to each item, so that subsequent modules can perform adaptive parameter correction and version traceability management based on the completion results, thereby realizing a state modeling foundation that is continuously synchronized with actual working conditions.

[0055] First, during the manufacturing phase before meter delivery, each meter undergoes a standardized factory calibration process. This process typically consists of multi-point load testing, high and low temperature testing, phase mismatch testing, and dynamic response measurement, resulting in a complete error profile encompassing amplitude response, phase offset, nonlinear error, and temperature drift behavior. Building on this foundation, the present invention structuredly extracts the archived calibration information for all meters to form a set of standard factory calibration parameters. This parameter set includes: static error curves describing frequency response and nonlinear error; a multi-point calibration gain matrix for each power level; offset values ​​for phase error correction based on varying load phase angles; and a drift function for the measured output under varying temperature. This function, represented as a mathematical model or sample data, is used to construct a characteristic temperature drift model. This set of parameters is used to construct an initial performance model, which describes the meter's theoretical measurement behavior under ideal operating conditions (i.e., standard ambient temperature, stable voltage, and minimal interference). This model provides a baseline response reference for subsequent state modeling.

[0056] Because meters may face varying environmental conditions at their actual deployment locations, their operational behavior is often influenced by climate characteristics, grid quality, equipment wiring methods, and load variations. Therefore, the present invention introduces an environmental adaptation mechanism to correct for drift in the initial performance model due to environmental differences. Specifically, the system correlates the deployment area code with a national or regional energy and environmental database to obtain key environmental characteristic variables such as the target area's annual average temperature and humidity range, typical grid voltage fluctuation amplitude, and peak-valley load curve shape. Taking the annual average temperature as an example, its value is used to correct the baseline parameters in the temperature drift model, ensuring that the temperature drift error term has localized characteristics at the initial modeling stage. The voltage fluctuation range is used to derive the transmission effect of input voltage disturbances on the stability of the measurement response, assisting in determining the dynamic drift error range. On this basis, the system constructs an environmental adaptation correction function that acts on each indicator of the initial performance model item by item, ensuring that the initial model has a certain degree of adaptability and practical reference value before observing the original data.

[0057] Next, to further tailor the device's actual operating characteristics during its current operating cycle, the present invention uses the raw measurement data sequence as a foundation and performs structured grouping by measurement channel (e.g., voltage channel, current channel, etc.). Within each channel's data, the system extracts statistical behavioral characteristics within a specific time period, including but not limited to the channel's maximum, minimum, average, standard deviation, instantaneous rate of change, zero-crossing period, and frequency domain dominant frequency characteristics. These characteristics are then compared with historical data from the meter's previous stable operating cycle, analyzing their range of variation, amplitude of fluctuation, and trend consistency. For example, if a channel exhibited a periodic increase alternating between day and night in historical operation but now exhibits a gentle fluctuation or even a reverse decline, this indicates that the channel may be experiencing dynamic drift or systematic errors in the current operating environment. Based on this, the system constructs a set of dynamic operating characteristics, each of which is labeled with a sampling interval and a confidence score to indicate whether the current measurement behavior deviates from the historical stable operating characteristics.

[0058] Based on the above dynamic operating characteristics and the modified initial performance model, the present invention adopts a strategy based on interval overlap optimization to integrate static calibration results with current behavioral characteristics to generate a structured set of state parameters. This set contains state boundary definitions in multiple dimensions, specifically including: a reference output range interval covering the measurement time axis, which is obtained by weighted extrapolation of the maximum and minimum values ​​over the past period of time; a trend direction boundary, which is used to determine whether the direction of data change is within the acceptable direction change range, and is particularly suitable for data sets with periodic data or slope-dominated characteristics; an upper limit on the change amplitude, which is used to calibrate the maximum change value per unit time and serve as a criterion for mutation detection; a tolerance correction factor, which is used to relax or tighten the error judgment boundary to adapt to load changes or signal noise levels; and a reasonable measurement interval definition, which is used to indicate the credible range in which the measurement value should be distributed under the current operating environment and exclude extreme values ​​that may be caused by abnormal factors such as harmonic interference, line coupling, and short-term disturbances.

[0059] The above-mentioned state parameter set is not a static configuration, but serves as a real-time reference for subsequent identification, judgment, prediction, and completion. In order to facilitate the dynamic management of the system, the present invention performs standardized packaging processing on the state parameter set, establishes a clear data structure and attaches version management information. Each parameter item is accompanied by a credibility label to identify whether the item comes from factory calibration, historical data modeling, or current operating cycle behavior extraction; it is also accompanied by an update timestamp and validity period range, which is used by the system module to determine whether the parameter item needs to be re-updated or temporarily frozen. The introduction of such metadata helps to dynamically adjust the reference boundaries of the state set according to the degree of match between the completed data and the current state parameters during the completion process, so that the modeling process has adaptive and backtracking capabilities, further enhancing the robustness and controllability of the system.

[0060] In addition, the generation of the state parameter set also provides a quantifiable and traceable basic framework for the entire completion chain. For example, in the incomplete identification stage, the deviation between the current observation value and the reference output interval can be compared to quickly determine whether it exceeds the tolerance threshold; in the completion stage, the trend boundary and the fluctuation amplitude can be combined to determine the reasonable completion interval range; in the post-completion state update stage, the change amplitude item and tolerance correction factor in the state parameter set will participate in the credibility assessment and scoring standard construction of the completion value, thereby completing the adaptive update and state synchronization mechanism. Therefore, the state parameter set not only undertakes the basic functions of data evaluation and anomaly screening, but also participates in decision support and result verification throughout the entire process, and is an indispensable core component for the normal operation of the system.

[0061] Step S103: Divide the original measurement data sequence into multiple data windows in chronological order. In each data window, combine the trend direction and change rate boundary recorded in the state parameter set to calculate the change amplitude and trend direction of the collected data in adjacent time periods, and predict the expected measurement value in the current window.

[0062] In step S103, the system's primary task is to segment the data into time segments based on the original measurement data sequence and predict the expected measurement value for the current period within each time window, thereby establishing a reasonable reference baseline for subsequent data anomaly identification and incomplete completion. Unlike traditional static valuation, this step not only considers the changing trend of the data itself during the prediction process, but also explicitly introduces dynamic information from the state parameter set, including trend direction, change rate boundaries, measurement response characteristics, etc., thereby achieving a dynamic prediction method that integrates historical behavior and current state, resulting in prediction results with higher accuracy and physical rationality.

[0063] Specifically, the system first partitions the raw measurement data sequence along the sampling timeline, forming a set of continuous and ordered time windows. Each time window can be set to a fixed length, such as 5 minutes, 15 minutes, or 1 hour, depending on the specific sampling period. During the partitioning process, the system allows for overlap between windows to ensure smooth transitions at the data boundaries, avoiding problems such as discontinuous predictions or sudden changes in responses. The measurement data within each time window serves as the input reference for the current cycle, used to predict the expected value at the next moment or the end of the current window.

[0064] Within each window, the system first extracts the difference in value change between adjacent sampling points, constructing a first-order difference sequence to quantify the magnitude and direction of change in the measured data within that window. For example, if the voltage measurement in a window rises from 223V to 227V, the difference is positive, and the trend direction is "upward." If it continuously decreases, it is "downward." If the fluctuations are small, they can be labeled "stable fluctuation" or "noise-dominated." This identification of trend direction provides a foundation for determining the direction of the forecast results.

[0065] In order to prevent prediction deviations caused by relying solely on local data, the system further introduces the trend direction and change rate boundaries in the state parameter set as adjustment factors. The state parameter set stores the typical behavior patterns of each measurement channel extracted from historical stable operation data. For example, a channel is usually in the peak load area between 10:00 and 12:00, with a fast current change rate and violent fluctuations; while at night, the voltage change of this channel is small and the fluctuation tends to be stable. The system will extract the corresponding reference trend direction and upper and lower limits of the change rate based on the timestamp and channel number of the current window to assist in determining whether the current differential result is within a reasonable range. If the actual change rate exceeds the preset boundary, the system will consider that the data segment may be affected by instantaneous disturbances, and will restrict its trend judgment during the forecast, or use the historical period average for smoothing.

[0066] After completing the trend direction and rate judgment, the system will perform the prediction operation of the expected measurement value in the current window. The prediction method can adopt local linear fitting, moving weighted average, sliding window regression and other methods. The selection should be based on the actual data structure and processing capabilities. During the calculation, the system first constructs a set of prediction variables, which usually includes: the values ​​of the previous to the first three sampling points, the differential slope, the mean offset of the channel in the state parameter set, the relative position of the prediction time in the 24-hour period, etc. Then a simple regression model is applied to weight these variables and output the predicted value. For example, suppose the first two sampling points are and , the slope is , then the predicted value can be expressed as ,in It is the trend amplification factor from the state parameter set, and its value usually ranges from 0.8 to 1.2 to adapt to the current operating status.

[0067] In order to enhance the robustness and flexibility of the prediction, the system also calibrates the predicted values. The calibration process includes upper and lower limit truncation and trend consistency comparison. Upper and lower limit truncation means that the system imposes physical boundary restrictions on the predicted value. For example, if the upper limit of a certain current is 100A, if the predicted value is 102A, it will be forced to be corrected to within 100A; trend consistency comparison means that the system compares the direction of change of the predicted value with the trend direction in the state parameter set. If the two do not match, for example, the trend is "downward" but the predicted value increases, the adjustment mechanism will be triggered to recalculate the trend slope or expand the window reference range to improve the stability of the prediction.

[0068] In special scenarios, such as unusually volatile data or strong interference from short-term sampling signals, the system enters a "weak prediction mode," prioritizing the historical mean of the state parameter set or the mean of the previous stable window to avoid outputting uncontrollable prediction results in the absence of credible input. After all predictions are completed, the system associates the predicted measurement values ​​within the current window with the timestamp and retains them in a cache queue, providing a reference for actual value comparison and incomplete judgment in the subsequent step S104.

[0069] Furthermore, the raw measurement data sequence is divided into a plurality of data windows in chronological order. Within each data window, the change amplitude and trend direction of the collected data in adjacent time periods are calculated in combination with the trend direction and change rate boundary recorded in the state parameter set, and the expected measurement value within the current window is predicted, including:

[0070] Based on the timestamp continuity of the original measurement data sequence, the complete sequence within the acquisition period is time-uniformly regularized, and overlapping data segments of equal length are generated according to a sliding window mechanism. Each data segment serves as an analysis window, where some historical sampling points are retained between adjacent windows to enhance context continuity.

[0071] In each analysis window, the short-term local change characteristics of the measurement data are extracted, including the first difference sequence, instantaneous change rate, local maximum and minimum fluctuation range, and matched with the trend direction and change rate boundary of the corresponding channel record in the state parameter set to calculate the trend deviation index of the window;

[0072] The mean, variance, and volatility of historical sampling points within the window are used as input features. Combined with the reference output range and tolerance factor under the current environment in the state parameter set, a local feature description vector is constructed. The support vector regression (SVR) model is then used to perform regression prediction on the expected value at the next time point within the window and output the predicted measurement value.

[0073] Compare the predicted value of this window with the predicted values ​​in the two adjacent windows before and after it. If there is a sudden deviation between the three that exceeds the trend smoothing boundary, the back-fitting operation is triggered and the regression weight is readjusted to improve the smoothness of the expected value.

[0074] The predicted measurement values ​​in all windows are paired with their corresponding sampling timestamps to form a predicted measurement sequence, and high-fluctuation windows, trend mutation points, and prediction confidence intervals are marked, providing a dynamic basis for the next step of deviation comparison and incomplete identification with the actual collected data.

[0075] During the continuous acquisition of measurement data, the most fundamental step is to time-warp the raw data series to ensure that all subsequent calculations are performed on a consistent time base. This warping process first verifies the timestamp consistency of the collected data. In a multi-source, parallel metering system, even if all devices are time-synchronized, timestamp errors or jumps may occur due to communication delays, buffer loss, or local storage latency. Therefore, the system reorders all sampled data in ascending order of their timestamps and removes artifacts caused by abnormal sampling intervals. For example, for a sequence with a nominal 200ms interval, if the interval between adjacent data points exceeds 600ms, it is considered to be missing or an abnormal jump, requiring manual interpolation or labeling. After warping, the continuous timeline is divided into multiple analysis windows, each containing data points of equal length. To enhance the prediction algorithm's understanding of context, each window is not completely independent. Instead, a sliding window mechanism is used, retaining a certain number of data points from the end of the previous window when generating the next window, thereby achieving overlap between data segments. This approach can effectively alleviate the noise sensitivity caused by mutation points to the model and improve the continuity and stability of the overall prediction.

[0076] The core task of each analysis window is to extract indicators that can represent trends and change characteristics from the measurement data within it. To achieve this goal, the system first performs differential processing on the data within the window to obtain a first-order difference sequence that reflects the speed of data change. This difference sequence is used to reflect the rapid fluctuation behavior of the data within the sampling time scale and is compared with the change rate boundary defined in the state parameter set to determine whether the window is currently within the normal fluctuation range. In addition, the system also calculates the local maximum and minimum values ​​in each window and uses this to construct a local fluctuation range to quantify the activity amplitude of the measurement signal in this time period. These indicators constitute the so-called short-term local change characteristics. The larger their value, the greater the instability of the current window data and the greater the possibility of deviation in the subsequent completion process.

[0077] The system further integrates statistical features such as the mean, variance, and volatility of all sampling points within each window into a set of input feature vectors. This is then fused with the channel's reference output range, trend direction, and tolerance factor, as recorded in the state parameter set. For example, if the mean of the current window deviates from the historical normal range but the difference is within the tolerance interval, the window can be considered acceptable; otherwise, additional smoothing constraints or exception handling mechanisms are required. These fused feature items are combined into a local feature description vector, which serves as the input to the prediction model.

[0078] Regarding the specific prediction method, the present invention preferably uses support vector regression (SVR) to calculate the expected measurement value at the next time point within a window. Specifically, based on historical sampling points within the current analysis window, statistical features such as mean, standard deviation, first-order difference, local fluctuation amplitude, and rate of change are extracted as input vectors. These features are then integrated with the trend direction, tolerance boundaries, and reference intervals defined in the state parameter set to form feature inputs for training and prediction. Next, multiple windows of known values ​​are selected from historical high-confidence data segments as training samples. A mapping relationship between the input features and the actual measurements is established. The kernel function used in the SVR method is used to perform feature space mapping and fit boundary optimization, ultimately determining the support vector set and regression coefficients. For the current prediction window, its feature vector is input into the trained SVR model, which outputs the corresponding predicted value for the next time point. This predicted value is then bound to its timestamp to form a predicted measurement sequence, which serves as a reference for subsequent incomplete identification and completion operations. This method exhibits excellent nonlinear modeling capabilities and boundary control, making it particularly suitable for scenarios where meter measurement data exhibits slight fluctuations but a continuous overall trend.

[0079] To ensure smoothness and verifiability of prediction results, this paper designs a back-fitting mechanism based on window sequence continuity judgment in conjunction with support vector regression. The core idea of ​​this mechanism is to leverage the characteristics of SVR in time series prediction: its sensitivity to input feature changes and its output being constrained by support vectors. By comparing the prediction results of adjacent windows, it dynamically corrects nodes with abnormal deviations in the prediction path, ensuring the trend consistency of the overall prediction curve.

[0080] Specifically, after completing the support vector regression prediction for the current window, the system calls the predicted measurements for the previous and next windows on the same channel, using the same SVR model structure or its retained state as the basis for prediction. Subsequently, the trend slope difference, incremental change, and prediction residual variance between the current window prediction and the two previous and next window predictions are calculated, and these results are compared with the trend change boundary, slope tolerance, and residual threshold defined in the state parameter set.

[0081] If any deviation between the current window's forecast and the adjacent window's forecast exceeds a set threshold, the system determines that the forecast point may have a sudden trend change or an inadequate regression fit. At this point, the system triggers a backfitting mechanism, retaining the original input features of the current window but modifying the components of its feature vector. This modification includes, but is not limited to, increasing the relative weight of trend direction features (such as first-order differences or historical slope averages), reducing the influence of high-volatility features (such as instantaneous maximum fluctuation amplitude or standard deviation), or adjusting the kernel function hyperparameters of the SVR regression, such as reducing the tolerance factor ε or adjusting the penalty coefficient C, to increase the sensitivity of the fit to outliers.

[0082] The backtracked, corrected feature vector is re-entered into the support vector regression calculation process to generate a new prediction value, which is then compared with the previous result. If the new prediction value is closer to the trend boundary midline than the previous one and no longer triggers the mutation condition, the revised prediction value is used as the final output. Otherwise, further rounds of backtracking are triggered, or if the confidence level is extremely low, the window is marked as unstable and handed over to the defect recognition module for processing as a potential anomaly.

[0083] The backtracking fitting mechanism described in the present invention combines the high-dimensional feature mapping and tolerance control capabilities of support vector regression, and realizes flexible control of prediction errors through trend linkage judgment among multiple windows and dynamic adjustment of local feature weights. It can not only improve the accuracy of incomplete identification before completion, but also effectively avoid the chain deviation caused by short-term anomalies on the global prediction path. It is suitable for the continuity reconstruction of power data and abnormal warning under complex working conditions.

[0084] After completing the above prediction steps, the system binds and pairs the predicted measurement values ​​in all analysis windows with their corresponding timestamps, generating a chronologically ordered sequence of predicted measurements. This sequence is not the final output used for completion, but serves as a key reference for subsequent steps. The identification results of high-volatility windows are used to guide the confidence assessment of incomplete data points, trend mutation points are used to assist in determining whether cross-region anomalies exist, and the predicted confidence interval provides numerical boundaries for determining whether it falls within the reference output range. These output structures play a supporting role in subsequent steps such as incomplete identification, value replacement, and state update, forming the intermediate judgment chain throughout the completion process.

[0085] Step S104: Compare the predicted measurement value with the actual collected measurement value, and based on the reference output range and tolerance interval set in the state parameter set, determine if the difference between the two exceeds the set threshold, then record the corresponding position as incomplete data to form an incomplete position index set.

[0086] During the data defect identification process, the predicted expected measurement values ​​in each data window need to be compared point by point with the actual collected original measurement values ​​to identify potential anomalies or data loss locations. In this step, the predicted measurement values ​​are calculated based on the data change trend of the previous period and the change pattern recorded in the state parameter set, while the actual measurement values ​​are derived from the actual measurement records of the meter itself. Both reflect the system's quantitative characterization of the difference between normal operation and actual operation. To ensure the scientific nature and controllability of the difference judgment, it is necessary to introduce multiple key reference quantities in the state parameter set, including the reference output range, the change trend tolerance interval, and the historical statistical deviation tolerance boundary.

[0087] Specifically, the state parameter set, initialized in the previous step, contains the following key fields: the reference output range consisting of the minimum and maximum values ​​of each measurement type (such as voltage, current, active power, and reactive power) under stable operating conditions in the current environment; the upper and lower limits of the permissible rate of change of the same measurement within a continuous time period, which constitute the change trend tolerance interval; and the standard deviation statistics generated based on historical operating samples, which are used to define the permissible transient disturbance fluctuation limit. In this step, the system first iterates over each sampling point using a data window alignment method, performing a numerical difference operation between the predicted measurement value at the current point and the actual measurement value, obtaining the difference Δ(t). The system then determines whether this difference Δ(t) exceeds the permissible error limit defined in the state parameter set for the current measurement. This limit can be a static set value (e.g., an absolute deviation of no more than 3%) or a dynamically calculated result (e.g., no more than 3 standard deviations) to improve adaptability.

[0088] To enhance robustness, the system not only uses single-point differences but also employs a moving window averaging method. This method calculates the root mean square error (RMSE) or sliding mean absolute error (SMAE) between the predicted and actual values ​​within a small window consisting of the current point and several adjacent points. This method mitigates misjudgments caused by occasional measurement spikes. If these difference metrics exceed the reference output range and tolerance interval, the system deems the data point potentially incomplete and records its location index in a "incomplete location index set." This index set, based on timestamps, can record multiple, continuous or discontinuous data incomplete points for subsequent access during completion.

[0089] During implementation, to ensure consistent applicability for different measurement quantities, the system will call corresponding tolerance models for different types of data (such as electrical parameters, meteorological quantities, and communication quality indicators) based on the classification settings in the state parameter set. For example, for current-related measurement quantities, the state parameter set generally sets a relative change tolerance (such as 5%); while for derivative parameters such as power factor, an absolute threshold may be set (such as the deviation must not exceed 0.05). At the same time, for measurement data in special scenarios (such as non-steady states caused by sudden load changes, equipment start-up and shutdown, etc.), the system can also temporarily suspend the incomplete judgment of the data segment based on the operating mode switching indicator marked in the state parameter set to prevent normal fluctuations from being mistakenly judged as abnormalities.

[0090] Ultimately, all data points deemed incomplete are aggregated into a chronologically ordered incomplete location index set, providing a clear basis for subsequent completion operations. This set includes not only the time index and measurement channel identifier for each incomplete point, but also metadata such as difference indicators and anomaly level scores, improving the accuracy and efficiency of the completion strategy.

[0091] Furthermore, the predicted measurement value is compared with the actual measurement value, and if the difference between the two exceeds a set threshold value based on the reference output range and tolerance interval set in the state parameter set, the corresponding position is recorded as incomplete data to form an incomplete position index set, including:

[0092] Based on the predicted measurement values ​​and the actual collected values ​​in each data window, a residual sequence is constructed, where each residual value is the difference between the predicted value and the actual value, and the constructed residual sequence is standardized so that it is easy to uniformly distinguish within the tolerance boundary set in the state parameter set;

[0093] According to the tolerance of measurement fluctuations under various working conditions in the state parameter set, a multi-level tolerance threshold structure is set, including absolute difference threshold, relative deviation percentage threshold, and dynamic threshold related to historical error distribution. A corresponding confidence annotation matrix is ​​established to preliminarily screen out over-threshold residual points as candidates for identification.

[0094] A contextual consistency check mechanism is introduced into the defective candidate points. The residual change rate of the N adjacent time points before and after each candidate point is calculated. If the residual fluctuation of the current point significantly deviates from the trend continuity expectation, that is, the breakpoint exceeds the limit, it is promoted to a high-confidence defective point and a dynamic defect marker list is established.

[0095] Combined with the historical behavior label information of similar meters in the state parameter set, including operating mode or environmental disturbance labels, the candidate defect points are scored using contextual semantic weighting, non-systematic deviations caused by the environment are eliminated, and the defect location index set is finally determined; wherein, the operating mode includes peak mode and standby mode; the environmental disturbance labels include temperature and humidity fluctuations and load mutations.

[0096] In an adaptive meter state estimation method that supports data incompleteness, to accurately identify possible missing or abnormal locations in a measurement data sequence, the predicted measurement value must first be compared point by point with the actual measured value. The difference between the two values ​​is then determined to determine whether it exceeds a threshold, based on the reference output range and tolerance interval defined in a preset set of state parameters. If the deviation at a location significantly exceeds the set tolerance, that location is identified as having data incompleteness, and a set of missing location indexes is generated for subsequent data completion.

[0097] In practice, this method segments the entire raw measurement data sequence using a sliding window mechanism. Each sliding window contains a certain number of consecutive sampling points, and the window length can be dynamically adjusted based on the sampling period, data change frequency, and expected processing delay. Within the window, the system first pairs the predicted measurement value with the actual collected value at each sampling moment and performs a subtraction operation on each pair to obtain the residual at that sampling moment. By traversing the entire window, a complete residual sequence is constructed, which is then used to determine whether the abnormality tolerance range is exceeded.

[0098] To avoid scale confusion caused by different types of measured physical quantities, inconsistent units, or differences in sensor ranges, the system standardizes each residual sequence. Specifically, standardization uses a unified normalization formula, based on the output reference range and tolerance factor preset for different measurement channels in the state parameter set, to perform a dimensionless transformation on the original residual. Each standardized residual value is calculated by dividing the original residual value by the maximum tolerance limit of the corresponding channel, normalizing it to the standard judgment interval between [-1, 1]. This ensures that in a multi-channel, multi-dimensional data environment, the system can use a unified standard to determine whether each sampling point has abnormal deviations.

[0099] After generating the standardized residual vector, the system further introduces a composite tolerance determination mechanism to refine the outlier identification logic. This mechanism primarily consists of three complementary threshold components: an absolute difference threshold, a relative deviation percentage threshold, and a dynamic tolerance threshold based on the historical residual distribution. The absolute difference threshold is based on the maximum allowable error set for each channel. For example, it can be set to ±5A for the current measurement channel and ±10V for the voltage channel. The relative deviation percentage threshold is the percentage difference between the predicted and actual values, accounting for relative errors with varying measurement amplitudes. Typical values ​​range from 5% to 10%. The dynamic tolerance threshold is based on an error distribution model constructed through analysis of historical operating data. It reflects the confidence interval for the residuals of a particular type of meter under similar operating conditions. For example, the mean and standard deviation are calculated based on historical data and set within the range of ±2σ of the mean. When determining whether a point is outlier, the system integrates these three threshold components. A point is marked as a preliminary outlier candidate only if the standardized residual simultaneously meets multiple out-of-limit conditions.

[0100] For initially selected candidate points, the system introduces a contextual consistency check mechanism to further determine whether the anomaly exhibits a break in temporal continuity. The basic concept behind contextual consistency checking is that, under normal conditions, meter data should exhibit a certain degree of temporal continuity and consistent trend. If a sudden change occurs at a point, it can be inferred to be an anomaly or measurement incompleteness. To this end, the system extends N sampling points before and after each candidate point to form a local time period and calculates the rate of change of the residuals for each sampling point within this time period. The rate of change is calculated by dividing the difference between adjacent residual values ​​by the time interval. During the analysis process, if the residual change rate of the current candidate point deviates significantly from that of the preceding and succeeding points, such as by more than two standard deviations of the mean rate of change, it is considered a trend break. This point is then promoted to a high-confidence defect point and recorded in the dynamic defect marker list, providing a basis for confidence weighting in the subsequent completion process.

[0101] Considering that measurement data may produce unstructured short-term deviations due to the influence of operating status or environmental disturbances, in order to prevent the system from misjudging such normal changes as missing, a historical operating behavior label and environmental disturbance label mechanism are introduced into the state parameter set to provide a basis for contextual semantic judgment. The operating behavior label describes the working state of the meter, including peak operation, standby, load regulation, power factor correction and other modes; the environmental disturbance label indicates whether there are external factors such as sudden changes in temperature and humidity, voltage fluctuations, and lightning interference at a specific moment that may affect measurement accuracy. The system matches the behavior label for each candidate residual point. If the residual deviation at the time of the point can be reasonably explained by the known behavior label, the system will reduce the abnormal confidence of the point to avoid misjudging it as a true missing point.

[0102] After normalizing all residual anomalies, verifying contextual continuity, and weighting behavioral semantic labels, the system will ultimately identify those sampling points that meet the residual threshold, meet multiple criteria, exhibit significant contextual discontinuity, and have unexplained semantic annotations as final missing locations. These points will form a missing location index set, which will be stored in system memory or persistently recorded as an index to guide subsequent data completion.

[0103] It is worth noting that the incomplete position index set is not just a list of abnormal positions, but also contains additional attributes such as the abnormality type, over-limit amplitude, trend change rate, semantic label weight, etc. corresponding to each position. This information will be used as a constraint or scoring factor in the weight calculation and selection judgment of the interpolation model during the completion process. For example, high-confidence residual points located in the trend break area will be reconstructed by calling the nonlinear trend modeling method first, while low-confidence residual points affected by external environmental disturbances can adopt a more conservative mean interpolation strategy to prevent the introduction of false data by mistake. This differentiated processing mechanism improves the robustness and accuracy of the overall completion and makes this method more adaptable when facing diverse power measurement data scenarios.

[0104] Step S105: For the incomplete position index set, by analyzing the numerical change trend and continuity relationship of the data collected before and after the incomplete position, and searching for the completion value within the reasonable measurement interval defined in the state parameter set, the completed measurement data sequence is estimated and generated.

[0105] After identifying the missing locations, a targeted completion process must be performed on the resulting incomplete location index set to generate a continuous and consistent measurement data sequence. The key to this completion process lies in both adhering to the data's inherent variation patterns and integrating it with the reasonable measurement intervals and variation trends defined by the previously constructed state parameter set. This ensures that the completed data is both authentic and reliable, and conforms to the meter's behavioral characteristics under specific environmental and load conditions.

[0106] When actually executing this step, first determine the valid data collected at the adjacent time points for each incomplete position. This process not only includes the data samples immediately before and after, but can also be appropriately extended to historical valid data points in a wider time window to enhance the robustness of trend inference. The system calculates the numerical difference, time interval, and growth or decay direction between these data points to form a short-term change path for the missing position, and then models its continuity in the time dimension. In the modeling process, the sliding weighted difference method or the first-order trend fitting method is preferably used to extract the main change gradient between the previous and next sampling points. If the direction of change is basically consistent and the gradient is within the rate range defined in the state parameter set, the trend is judged to be stable and suitable for linear extrapolation; if the previous and next data show fluctuation characteristics or the rate exceeds the allowable range, the segmented trend approximation or the two-way clamping method based on the multi-value upper and lower limit envelope is used to improve the accuracy of the upper and lower bounds of the prediction.

[0107] During the search for a completed value, the system first generates a draft estimate based on a trend model. Using this estimate as the center, the system then performs a constraint check within the reasonable measurement range recorded in the state parameter set. This reasonable measurement range is typically constructed based on the statistical distribution range of historical operating data, the maximum allowable error band defined by manufacturing calibration parameters, and typical numerical variation trajectories under similar operating conditions. Specifically, each type of measurement variable is assigned an upper and lower limit, as well as a desired rate of change boundary. For example, for a voltage parameter in a 220V system, the reasonable range can be defined as between 210V and 230V. Current parameters are set from 0 to a certain maximum current threshold based on the load characteristics and dynamically adjusted with load changes. If the estimated value falls within this reasonable range, it is directly written into the completion sequence as the completed value. If it exceeds either the upper or lower limit of the range, the closest legal boundary value to the estimated value is selected as the completed result based on the upper and lower bounds principle, ensuring that the result neither violates system stability nor introduces abnormal fluctuations.

[0108] During the search for completion values, an auxiliary judgment mechanism based on the "context consistency" of the incomplete position is also introduced. This mechanism evaluates whether the change pattern of the completion position in its time window is consistent with the previous and subsequent continuous regions. In other words, it examines whether the trend consistency score and the fluctuation difference threshold meet the continuity rules defined in the state parameter set. If they are inconsistent, it means that the point may be in a jump segment or a data segment affected by interference. In this case, the completion process needs to be controlled by a more fault-tolerant strategy, such as global statistical regression analysis based on multiple historical windows, to avoid the propagation of completion errors caused by local anomalies.

[0109] Furthermore, to ensure the accuracy and adaptability of the completion process, the system will call upon information on the incomplete detection threshold and measurement-sensitive sections in the state parameter set in real time and match it with the actual data scenario to determine whether active interpolation, trend extrapolation, or boundary substitution are currently appropriate for completion. For example, if the state parameter set indicates that the current variable is in a "high dynamic fluctuation period" or a "measurement error-sensitive period," the system will prioritize a conservative strategy, implementing completion only when the confidence interval of the completed value is significantly better than the null value state. Otherwise, the missing mark will be retained to prevent the introduction of false information that may affect subsequent evaluations.

[0110] Finally, multiple incomplete locations are completed sequentially using the above method to form a completed measurement data sequence. To further improve data consistency and time series integrity, after the completion process is completed, the connection area between the completed segment and the original valid segment needs to be smoothed to eliminate possible abrupt seams. This smoothing process usually uses sliding weighted average, spline connection, or local reconstruction interpolation to ensure that the data sequence is continuous in the time dimension, physically reasonable, and statistically consistent with the system state parameters, thereby laying a stable and reliable foundation for subsequent state parameter set updates, adaptive adjustments, and confidence scoring operations.

[0111] Furthermore, for the incomplete position index set, by analyzing the numerical change trend and continuity relationship of the data collected before and after the incomplete position, and searching for the supplementary value within the reasonable measurement interval defined in the state parameter set, estimating and generating the supplemented measurement data sequence, the method includes:

[0112] Extract the time index corresponding to each incomplete position in the original measurement data sequence, and locate several consecutive collected measurement values ​​in front and behind it to form the front and back data segments;

[0113] Calculate the change amplitude between each pair of adjacent sampling points in the front and back data segments, and calculate their average change rate and change direction respectively;

[0114] Determine the type of change trend of the current defective position by combining the historical change range and speed boundary of the measurement data of this type recorded in the state parameter set under similar change trends;

[0115] According to the numerical difference between the preceding and succeeding data, a plurality of complementary value candidates are generated within the measurement allowable interval defined in the state parameter set by using an equidistant linear interpolation method or a trend extrapolation method;

[0116] For each candidate value, calculate the numerical deviation between it and the end value of the previous segment and the starting value of the next segment, and determine whether it meets the change rate boundary and tolerance requirements specified in the state parameter set at the same time;

[0117] The one with the smallest numerical deviation is selected from all the completion value candidates that meet the requirements as the final completion value, and is written into the corresponding incomplete position of the original measurement data sequence to estimate and generate the completed measurement data sequence.

[0118] In the proposed adaptive meter state estimation method supporting data incompleteness, the core goal of the completion process is to restore data at identified incomplete locations, ensuring that they are as close as possible to the actual measured values ​​under those conditions, thereby ensuring the integrity and continuity of the overall measurement sequence. To this end, the system accurately identifies the incomplete location index set and further designs a completion estimation process that combines local data trend analysis, state parameter set constraints, and candidate value screening mechanisms. This ensures that the completion results are both mathematically consistent and meet physical and environmental logic requirements.

[0119] In actual operation, the system first processes the identified incomplete position index set one by one. For each incomplete position, the system extracts its corresponding time index in the original measurement data sequence, which serves as a reference for all subsequent completion calculations. Based on this time index, the system searches forward and backward for a segment of completely recorded continuous sampling point data, forming the "front data segment" and "back data segment" respectively. The length of these two data segments can be set based on the measurement frequency, data change rate, and recommended values ​​in the state parameter set. It is generally recommended to be 3 to 10 data points to ensure sufficient contextual information and observability of local trends.

[0120] After obtaining the preceding and following data segments, the system then pairs adjacent sampling points within the segments and calculates the magnitude of change between them. By averaging the magnitude of change across all adjacent points within each segment, the average rate and direction of change for the preceding segment can be determined, and similarly for the following segment. This operation is significant because, through the approximate calculation of local derivatives, it captures the short-term trend of change in the surroundings before and after the current residual point, providing a foundation for trend classification and candidate generation.

[0121] Next, the system compares the calculated rate and direction of change for the preceding and succeeding segments with the historical change patterns recorded in the state parameter set for this type of measurement data. The state parameter set typically covers the boundaries of the rate of change, directional frequency distribution, and range of fluctuations that may occur in the measurement data per unit time under various operating modes, environmental conditions, and equipment states. This information, derived from historical equipment data analysis and engineering experience, is stable and instructive. Through this comparison, the system can determine which type of change trend the current defect is more likely to be in, such as a slow linear rise, periodic fluctuations, a climbing phase before a sharp jump, or a slightly disturbed interval within a stable state.

[0122] Once the trend type is preliminarily determined, the system proceeds to generate candidate values ​​based on the difference between the end value of the preceding segment and the start value of the following segment, combined with the identified trend characteristics. This stage is the core of the completion process, and its design must consider both mathematical feasibility and strict constraints on the reasonable measurement interval defined in the state parameter set. To this end, the system uses two methods to generate candidate values. The first is equidistant linear interpolation, which linearly distributes the data between the two segments, generating multiple intermediate values ​​that are numerically continuous but with fixed increments. The second is trend extrapolation, which uses the average rate and direction of change of the preceding or following segment to extend forward or backward along the time axis to predict several values ​​that fit the current trend. Each of these methods is suitable for different scenarios: the former is suitable for areas with gentle changes, while the latter is suitable for areas with sustained upward or downward trends. The system automatically selects or combines these methods based on the trend type.

[0123] After the candidate values ​​are generated, the system will calculate the difference between each candidate value and its adjacent previous end value and next start value to obtain its deviation measure within the context. This deviation is not a simple numerical difference, but needs to be judged in combination with the constraints such as the change rate boundary and error tolerance interval set in the state parameter set. For example, if the rate obtained by dividing the difference between a candidate value and the previous end value by the time interval exceeds the upper limit of change set in the historical record, even if the value is highly consistent with the subsequent data, it will be judged as an unreasonable completion value. Similarly, if the deviation value exceeds the allowable tolerance range, it will also be excluded from the candidate set. This process ensures that the completion value not only "fills the gap" in terms of numerical value, but is also an acceptable and reasonable understanding in terms of physical logic and historical experience.

[0124] After this rigorous screening, the system selects the candidate value that still meets the constraints and minimizes the combined deviation from the preceding and following data as the final completed value. This value is written to the corresponding missing position in the original measurement data sequence, generating an updated measurement sequence containing the completed value. It is worth noting that before writing, the system also adds a specific mark to the completed point to facilitate identification of the completed process during subsequent state estimation or data backtracking, further supporting parameter optimization and anomaly review.

[0125] The entire completion process not only strictly relies on the contextual data of the defect itself, but also deeply integrates the trend characteristics, change rate, and tolerance structure provided by the state parameter set, avoiding the static interpolation problem of "using points to represent the whole" common in traditional methods. Through this mechanism, the system can adaptively handle defect situations in different environments, different sampling frequencies, and different operating states, significantly improving the engineering usability and reasoning consistency of data completion.

[0126] Furthermore, in specific scenarios, the system can also choose to calculate the final completion value by taking a weighted average of multiple candidate values, where the weight coefficients are set inversely proportional to the difference between the front-end and back-end values. While this approach sacrifices certainty, it can effectively mitigate mutations and improve overall sequence smoothness in situations with high noise, unclear contextual trends, or a large number of completion points, demonstrating the flexibility and robustness of the present invention's completion strategy.

[0127] Step S106: During the completion process, the state parameter set is corrected and adaptively updated according to the offset between the completed value and the state parameter set, so as to output the final completed measurement data sequence, the updated state parameter set and the incompleteness confidence score.

[0128] After completing the initial completion of incomplete measurement points, to further improve the stability, adaptability, and prediction accuracy of the entire estimation method during actual operation, the system needs to dynamically modify the originally constructed state parameter set based on the feedback from the various results during the completion process, and output the modified complete measurement data sequence, the updated state parameter set, and the incomplete confidence score used to quantify the completion quality. The state parameter set has been constructed in the previous step by combining the original measurement data sequence, manufacturing calibration parameters, and historical stable operation data. It contains statistical and dynamic characteristics of multiple dimensions describing the meter measurement output under normal operating conditions, including but not limited to: measurement value reference range, change trend direction, rate boundary, reasonable measurement interval, tolerance boundary, short-term variability, and trend fluctuation confidence factor.

[0129] During implementation, the system first performs a directional analysis of each completed point based on the completed measurement data sequence obtained in step S105. Specifically, it quantifies the degree of deviation between the completed value at that point and its preset trend prediction value, average level value, and median value of the reasonable measurement interval in the state parameter set. This deviation analysis not only involves calculating numerical differences but also determining directional matching, rate consistency, and degree of variation. By calculating overall deviation statistics (such as average deviation, maximum dispersion, and directional consistency) for multiple completed points, the system determines whether the original state parameter set has significant deviation under the current operating conditions and whether it still accurately reflects the normal output behavior of the meter in the current operating environment.

[0130] If the offset values ​​of multiple completion points are detected to have consistent directionality (e.g., an overall bias toward the upper or lower boundary), this indicates that the original reference output range may have undergone a systematic shift. In this case, the system fine-tunes the reference output range and reasonable measurement interval boundaries defined in the state parameter set based on the concentrated distribution characteristics of the completion values. This adjustment method preferentially employs a sliding window approach for dynamic weighted regression. This involves calculating new upper and lower boundaries using a weighted average over multiple adjacent completion cycles, with the weights positively correlated with the confidence level of each completion point. The adjustment results are directly written into the state parameter set, constraining the next round of data prediction and defect identification.

[0131] On the other hand, if the trend direction of the completed point shows a significant deviation from the actual measurement points before and after it, the system triggers a trend update mechanism to reconstruct the trend direction vector and change rate boundary recorded in the state parameter set. The trend reconstruction process combines the new data stream formed by the actual sampling points and the completed points in the current working cycle, and uses weighted least squares or Kalman filtering algorithms to approximate the current trend path. Based on this, a revised trend expression is generated, thereby improving the responsiveness and fitting accuracy of the subsequent prediction model.

[0132] To further enhance the robustness and adaptability of this correction mechanism, the system introduces a field-by-field confidence factor control mechanism for the state parameter set. Specifically, for each parameter in the state parameter set, such as the upper and lower limits of the output range, trend boundaries, and rate thresholds, a dynamic confidence factor is defined to indicate the frequency of corrections over the last N cycles, directional consistency, and sensitivity to forecast errors. When the confidence factor falls below a certain threshold, the system triggers a static lock for that parameter, preventing drastic fluctuations in the state parameter set due to short-term anomalies and ensuring system stability and algorithm convergence.

[0133] After completing the correction of the state parameter set, the system will output a complete, continuous, and highly reliable measurement data sequence based on the fusion of the final supplementary data and the original measurement data. This data sequence serves as the final estimation result and can be used for subsequent advanced analysis tasks such as energy efficiency evaluation, power quality monitoring, or load trend modeling.

[0134] At the same time, in order to make the credibility of the completed data quantifiable and comparable, the system also introduces a generation mechanism for incomplete confidence scores in this step. The scoring mechanism is based on the joint modeling of multiple dimensional indicators, including but not limited to: (1) the position offset ratio between the completed value and the reference range in the state parameter set; (2) the degree of consistency of the projection of the completed value on the trend vector path; (3) whether the state parameter set is adjusted during the completion process; (4) the overall fluctuation intensity within the window where the completion point is located. The above indicators are input into the scoring function model after standardization, and the output value range is [0,1], where the closer the value is to 1, the higher the confidence.

[0135] Ultimately, the system outputs the scoring results, along with the updated state parameter set and completed measurement data, to form a complete estimation result package. This result not only supports self-verification and dynamic regulation of meter data quality, but also provides a highly reliable input basis for external energy management systems, smart grid control systems, or data cleaning modules, enabling a closed-loop control path for adaptive meter state estimation driven by both data ontology and environmental perception.

[0136] Furthermore, during the completion process, the state parameter set is corrected and adaptively updated according to the offset between the completed value and the state parameter set to output the final completed measurement data sequence, the updated state parameter set, and the incompleteness confidence score, including:

[0137] For each completion position, extract the corresponding change trend boundary and reference output interval in the state parameter set, and construct the trend offset and interval offset corresponding to the current completion value;

[0138] Calculating the influence coefficient of each completion position on the trend parameter and boundary parameter in the original state parameter set during the completion process according to the trend offset and the interval offset;

[0139] Using the influence coefficient, weighted correction is performed on the trend change rate, output fluctuation range, and change direction in the state parameter set to form a preliminary draft of the updated state parameter set;

[0140] Based on the updated state parameter set, the adaptability of all completed positions in the current interval is re-evaluated, and the degree of fit of each completed value relative to the trend boundary and the residual square value are calculated;

[0141] The degree of fit and the squared residual value of each completed position are normalized and summarized to generate a complete confidence score for the entire completion process, and the final completed measurement data sequence, the updated state parameter set, and the score result are output.

[0142] In the proposed adaptive meter state estimation method for supporting incomplete data completion, to further enhance the credibility and dynamic adaptability of the completed data, it is necessary not only to accurately complete the incomplete data locations but also to fully consider the offset relationship between the completed values ​​and the original state parameter set during the completion process, thereby dynamically updating the parameter set structure and achieving self-correction and evolutionary optimization of system parameters. This step introduces trend offset, interval offset, influence coefficient, fit, and residual quantization mechanisms to construct a closed-loop feedback structure that connects prediction, completion, and state learning. This ensures that the final output not only has data continuity but also reflects the evolutionary ability of system state cognition.

[0143] Specifically, after completing the completion process, the system first retrospectively analyzes each completed data location. It first identifies the reference information corresponding to the completed location in the state parameter set, including but not limited to the trend change rate boundaries (e.g., maximum allowable growth rate, minimum allowable decline rate) for the channel or measurement type under the current operating conditions, the output reference range (e.g., reasonable range boundaries for voltage, current, temperature, and humidity), and the fluctuation tolerance range. These parameters are typically determined during device initialization, long-term data learning, or engineering setup, and serve as the benchmark for judging system operational stability. After extraction, the system compares the completed data value against these parameters item by item to construct the "trend offset" and "interval offset" for the completed point. The trend offset is defined as the difference between the actual rate of change between the completed value and its previous sample value and the reference trend change rate; the interval offset is the deviation between the completed value and the midpoint of the set reference interval, measuring its deviation from the center position.

[0144] After obtaining the trend offsets and interval offsets of all completion points, the system further calculates the degree of influence on the trend parameters and boundary parameters in the original state parameter set based on these values, and formally constructs an "influence coefficient." The construction of the influence coefficient not only considers the absolute value of the offset, but also considers comprehensive factors such as its position weight in the entire completion process, the stability of the historical data on which the completion value depends, and the severity of the fluctuations in the current working conditions. For example, if a completion value appears in a period of high volatility, it itself has a certain degree of prediction uncertainty, and its influence weight on the state parameter can be reduced accordingly; conversely, if its context trend is extremely stable and the completion value deviation is significant, it can be regarded as a mismatch in the state model, and the correction weight of the parameter at this point should be significantly increased. The design of the influence coefficient is essentially a robust weighting mechanism that ensures the system's sensitive response to the completion process while avoiding overall state drift due to single-point anomalies.

[0145] Through the calculation of the above-mentioned influence coefficients, the system begins to perform weighted corrections on relevant trend change rates, output fluctuation ranges, change directions and other factors in the state parameter set, and obtains a preliminary draft of an updated state parameter set. The correction strategy does not simply replace the original parameters, but rather introduces the feedback effect of the new complementary values ​​into the original parameters in a moderate manner by means of weighted averaging, offset compensation, dynamic compression, etc. For example, if the trend offsets of multiple complementary values ​​are concentrated in a certain direction, it means that the original trend change rate boundary setting is too conservative or deviates from the current operation. In this case, the change rate boundary needs to be appropriately expanded to cover the true trend; for example, if multiple complementary values ​​deviate from the center of the reference interval, it means that the output reference interval should be offset or recalibrated to adapt to the actual output changes of the current sensor or measured object.

[0146] After the initial correction, in order to ensure the rationality and effectiveness of the updated parameter set, the system also needs to re-evaluate the adaptability of all completed positions under the current parameters. This process is the internal consistency test of the completion results. Its core lies in recalculating the degree of fit and the residual square value of each completed point under the updated parameter set. The degree of fit can be obtained by calculating the weighted sum of multi-dimensional indicators such as the projected distance of the completed value relative to its context trend path, the slope fit to the reference trend, and the normalized offset to the center of the interval; the residual square value directly measures the numerical difference between the completed value and its context fitting trend, reflecting the confidence basis of the completed point. In this step, the system will use a combination of original historical data, neighborhood stability indicators and parameter mapping mechanisms to ensure that the evaluation of each completed point is based not only on local data, but also on its structural position and semantic attributes in the entire measurement sequence.

[0147] After completing this reassessment, the system normalizes the fit metrics and squared residuals of all completed points to eliminate dimensionality and outlier effects. These metrics are then aggregated to generate a "defect confidence score" for the entire completion process. This score is a comprehensive evaluation metric that reflects the stability, accuracy, and system consistency of the entire completion process within the current state cognition system. A higher score indicates greater synergy between the predicted values ​​of the completion process and the state model, and a more reliable system. A lower score indicates the need to further incorporate external information or extend the window depth to correct for model drift.

[0148] This paper further designs a defect confidence scoring mechanism to systematically measure the comprehensive performance of the completion process in terms of trend alignment, data consistency, and model drift tolerance. This scoring mechanism not only serves as a credibility indicator for the results but also provides input for subsequent dynamic adjustments and anomaly warnings.

[0149] Specifically, the system first extracts two types of evaluation indicators for each completed position: one is the trend fit indicator, which reflects the consistency of the completed value within the context trend; the other is the residual squared value indicator, which reflects the numerical error of the completed value relative to the prediction model path. Trend fit is calculated by calculating the angle between the completed value and the trend vector formed by its previous and next sampling points. A smaller angle indicates a consistent trend direction and a higher degree of fit. The residual squared value is calculated using the traditional squared error method, squaring the difference between the completed value and the prediction model output value as an error measure.

[0150] The system then normalizes the trend fit and squared residuals for all completed locations. Trend fit is converted to a confidence weight within the [0, 1] interval through maximum normalization, while squared residuals are normalized using the standard deviation of the residual distribution as the denominator to generate a uniformly scaled error term. These two values ​​are weighted and combined using a pre-defined weighting factor (e.g., 60% for trend fit and 40% for residual error) to form a local confidence score for each completed point.

[0151] Finally, the system performs a weighted summation of the local confidence scores for all completed points and normalizes this score based on the proportion of incomplete locations in the entire sequence, resulting in a score between [0 and 1]. The closer this score is to 1, the better the completion process performs under the current state parameter settings, with higher data consistency and more complete trend matching. Conversely, a lower score indicates a structural deviation between the state parameter model and the current data, potentially requiring parameter reinitialization or adjustment of the completion logic.

[0152] Ultimately, the system outputs three core results: 1. The final completed measurement data sequence after all completion operations are completed, in which all replaced data points already contain composite logic support from trend prediction, context matching and state mapping; 2. The updated state parameter set, which is corrected through feedback from multiple completion positions and represents the optimal estimate of the current working conditions and measurement system status, with strong timeliness and adaptability; 3. The incomplete confidence score, which can be used as input for subsequent exception handling, system reliability assessment, data quality control and exception alarms, and supports more advanced system governance.

[0153] Overall, the adaptive update mechanism constructed in this step is a closed-loop feedback system that realizes the full process linkage from data completion to state correction to credibility assessment. It not only improves the accuracy of the completion results, but also enhances the system's evolutionary learning ability.

[0154] A second embodiment of the application provides an electronic device, comprising:

[0155] processor;

[0156] The memory is used to store a program. When the program is read and executed by the processor, the program executes a meter state adaptive estimation method supporting data completion provided in the first embodiment of the present application.

[0157] The third embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for adaptively estimating a meter state that supports data incomplete completion provided in the first embodiment of the present application is executed.

[0158] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A method for adaptively estimating meter status supporting data incompleteness, characterized in that: include: Under a unified time base, the continuous measurement data of the meter is collected to form the original measurement data sequence; Based on the original measurement data sequence, combined with calibration parameters in the meter manufacturing process and pre-collected historical stable operation data, a state parameter set is generated, wherein the state parameter set is used to describe the normal output range and change trend of the meter under the current working environment; Dividing the original measurement data sequence into multiple data windows in chronological order, within each data window, combining the trend direction and change rate boundary recorded in the state parameter set, calculating the change amplitude and trend direction of the collected data in adjacent time periods, and predicting the expected measurement value in the current window; Comparing the predicted measurement value with the actual collected measurement value, and judging, based on the reference output range and tolerance interval set in the state parameter set, if the difference between the two exceeds a set threshold, recording the corresponding position as incomplete data to form an incomplete position index set; For the incomplete position index set, by analyzing the numerical change trend and continuity relationship of the data collected before and after the incomplete position, and searching for the supplementary value within the reasonable measurement interval defined in the state parameter set, the supplemented measurement data sequence is estimated and generated; During the completion process, the state parameter set is corrected and adaptively updated according to the offset between the completed value and the state parameter set to output the final completed measurement data sequence, the updated state parameter set and the incomplete confidence score.

2. The meter state adaptive estimation method supporting data incomplete completion according to claim 1 is characterized in that: The method of collecting the continuous measurement data of the meter under the unified time reference to form the original measurement data sequence includes: A unified time synchronization device is embedded in each meter. The device maintains local time based on a high-stability temperature-compensated crystal oscillator and aligns clocks via a time beacon periodically broadcast by an external master control node, ensuring global consistency of data collected from different meters. After all meters are synchronized, continuous measurement data is collected in parallel from multiple channels including voltage, current, active power, reactive power, power factor, and frequency based on the set minimum sampling period. A corresponding timestamp tag is added to each channel to generate the initial measurement data fragment. The initial measurement data fragments generated by each meter are transmitted to the edge node. Based on the local cache mechanism and timestamp aggregation algorithm, the edge node performs window alignment and format standardization on the different channel data from multiple meters according to a unified time base, forming a structured unified measurement data set. Marking abnormal values, missing values, or mutation values ​​that cross boundaries in the unified measurement data set using data quality screening rules, calculating confidence scores based on data distribution density within a window, and dividing data segments into high-confidence segments and low-confidence segments; High-confidence segments are stored as the backbone of the original measurement data sequence, and low-confidence segments are marked as key monitoring areas in the subsequent state modeling and incomplete recognition process. At the same time, corresponding confidence mask vectors are generated as important references in the subsequent prediction and completion links.

3. The method for adaptively estimating meter status with support for data incompleteness completion according to claim 1, characterized in that: The state parameter set is generated based on the original measurement data sequence and in combination with calibration parameters in the meter manufacturing process and pre-stated historical stable operation data, including: Extracting archived factory calibration parameter sets from the meter manufacturing process, including static error curves, multi-point calibration gain coefficients, phase correction offset values, and characteristic temperature drift models, to construct an initial performance model of the meter. The initial performance model describes the basic response characteristics of the measurement output under ideal operating conditions. Mapping and matching the initial performance model with the operating environment conditions of the meter's current deployment location, obtaining environmental characteristic variables including annual average temperature, humidity, grid voltage fluctuation range, and typical load curve by calling the environmental database of the deployment area, constructing an environmental adaptation model, and correcting drift terms in the initial performance model that may be caused by environmental changes; The collected raw measurement data series are grouped according to different data channels. Based on the fluctuation range, change trend and frequency domain response characteristics of each channel during the historical stable operation period, statistical stability indicators and change rate indicators are extracted to construct a dynamic operation feature set for actual operating conditions. Based on the dynamic operation feature set, an interval overlapping optimization algorithm is used to jointly calibrate the initial performance model and the fluctuation characteristics in the current operation data to generate a reference output range interval covering the time series, trend direction boundary, change amplitude upper limit, tolerance correction factor and reasonable measurement interval definition, thereby forming a structured state parameter set; The generated state parameter set is encapsulated in a standardized manner, and a credibility identifier and update timestamp are attached to each item, so that subsequent modules can perform adaptive parameter correction and version traceability management based on the completion results, thereby realizing a state modeling foundation that is continuously synchronized with actual working conditions.

4. The method for adaptively estimating meter status supporting data incompleteness and completion according to claim 1, characterized in that: The original measurement data sequence is divided into a plurality of data windows in chronological order. Within each data window, the change amplitude and trend direction of the collected data in adjacent time periods are calculated in combination with the trend direction and change rate boundary recorded in the state parameter set, and the expected measurement value within the current window is predicted, including: Based on the timestamp continuity of the original measurement data sequence, the complete sequence within the acquisition period is time-uniformly regularized, and overlapping data segments of equal length are generated according to a sliding window mechanism. Each data segment serves as an analysis window, where some historical sampling points are retained between adjacent windows to enhance context continuity. In each analysis window, the short-term local change characteristics of the measurement data are extracted, including the first difference sequence, instantaneous change rate, local maximum and minimum fluctuation range, and matched with the trend direction and change rate boundary of the corresponding channel record in the state parameter set to calculate the trend deviation index of the window; The mean, variance, and volatility of historical sampling points within the window are used as input features. Combined with the reference output range and tolerance factor under the current environment in the state parameter set, a local feature description vector is constructed. The support vector regression (SVR) model is then used to perform regression prediction on the expected value at the next time point within the window and output the predicted measurement value. Compare the predicted value of this window with the predicted values ​​in the two adjacent windows before and after it. If there is a sudden deviation between the three that exceeds the trend smoothing boundary, the back-fitting operation is triggered and the regression weight is readjusted to improve the smoothness of the expected value. The predicted measurement values ​​in all windows are paired with their corresponding sampling timestamps to form a predicted measurement sequence, and high-fluctuation windows, trend mutation points, and prediction confidence intervals are marked, providing a dynamic basis for the next step of deviation comparison and incomplete identification with the actual collected data.

5. The meter state adaptive estimation method supporting data incomplete completion according to claim 1, characterized in that: The predicted measurement value is compared with the actual measurement value, and if the difference between the two exceeds the set threshold value based on the reference output range and tolerance interval set in the state parameter set, the corresponding position is recorded as incomplete data to form an incomplete position index set, including: Based on the predicted measurement values ​​and the actual collected values ​​in each data window, a residual sequence is constructed, where each residual value is the difference between the predicted value and the actual value, and the constructed residual sequence is standardized so that it is easy to uniformly distinguish within the tolerance boundary set in the state parameter set; According to the tolerance of measurement fluctuations under various working conditions in the state parameter set, a multi-level tolerance threshold structure is set, including absolute difference threshold, relative deviation percentage threshold, and dynamic threshold related to historical error distribution. A corresponding confidence annotation matrix is ​​established to preliminarily screen out over-threshold residual points as candidates for identification. A contextual consistency check mechanism is introduced into the defective candidate points. The residual change rate of the N adjacent time points before and after each candidate point is calculated. If the residual fluctuation of the current point significantly deviates from the trend continuity expectation, that is, the breakpoint exceeds the limit, it is promoted to a high-confidence defective point and a dynamic defect marker list is established. Combined with the historical behavior label information of similar meters in the state parameter set, including operating mode or environmental disturbance labels, the candidate defect points are scored using contextual semantic weighting, non-systematic deviations caused by the environment are eliminated, and the defect location index set is finally determined; wherein, the operating mode includes peak mode and standby mode; the environmental disturbance labels include temperature and humidity fluctuations and load mutations.

6. The method for adaptively estimating meter status supporting data incompleteness and completion according to claim 1, characterized in that: For the incomplete position index set, by analyzing the numerical change trend and continuity relationship of the data collected before and after the incomplete position, searching for a completion value within a reasonable measurement interval defined in the state parameter set, and estimating and generating a completed measurement data sequence, the method includes: Extract the time index corresponding to each incomplete position in the original measurement data sequence, and locate several consecutive collected measurement values ​​in front and behind it to form the front and back data segments; Calculate the change amplitude between each pair of adjacent sampling points in the front and back data segments, and calculate their average change rate and change direction respectively; Determine the type of change trend of the current defective position by combining the historical change range and speed boundary of the measurement data of this type recorded in the state parameter set under similar change trends; According to the numerical difference between the preceding and succeeding data, a plurality of complementary value candidates are generated within the measurement allowable interval defined in the state parameter set by using an equidistant linear interpolation method or a trend extrapolation method; For each candidate value, calculate the numerical deviation between it and the end value of the previous segment and the starting value of the next segment, and determine whether it meets the change rate boundary and tolerance requirements specified in the state parameter set at the same time; The one with the smallest numerical deviation is selected from all the completion value candidates that meet the requirements as the final completion value, and is written into the corresponding incomplete position of the original measurement data sequence to estimate and generate the completed measurement data sequence.

7. The method for adaptively estimating meter status supporting data incompleteness and completion according to claim 1, characterized in that: During the completion process, the state parameter set is corrected and adaptively updated according to the offset between the completed value and the state parameter set to output the final completed measurement data sequence, the updated state parameter set, and the incompleteness confidence score, including: For each completion position, extract the corresponding change trend boundary and reference output interval in the state parameter set, and construct the trend offset and interval offset corresponding to the current completion value; Calculating the influence coefficient of each completion position on the trend parameter and boundary parameter in the original state parameter set during the completion process according to the trend offset and the interval offset; Using the influence coefficient, weighted correction is performed on the trend change rate, output fluctuation range, and change direction in the state parameter set to form a preliminary draft of the updated state parameter set; Based on the updated state parameter set, the adaptability of all completed positions in the current interval is re-evaluated, and the degree of fit of each completed value relative to the trend boundary and the residual square value are calculated; The degree of fit and the squared residual value of each completed position are normalized and summarized to generate a complete confidence score for the entire completion process, and the final completed measurement data sequence, the updated state parameter set, and the score result are output.

Images