Intelligent compensation method and system for error of electric energy metering box based on time sequence anomaly detection

Abstract

This invention discloses an intelligent compensation method and system for electricity metering box errors based on time-series anomaly detection, relating to the field of electricity metering error correction. The method includes: acquiring multi-dimensional time-series operational data streams; combining historical operational data from the electricity metering box to segment and filter initial abnormal operational data segments; configuring dual-path parallel hardware verification logic circuits, with the first path connected to the sampling signal of the main metering circuit and the second path connected to the synchronization reference signal of adjacent metering boxes in the same area; using the hardware verification logic circuits to perform cross-verification on the initial abnormal operational data segments to obtain the true abnormal operational data segments; introducing a joint search mechanism to perform a root cause search for metering errors on the true abnormal operational data segments, determining the target root cause of the metering error, and performing multi-scale dynamic error compensation. This application solves the technical problem of existing electricity metering box error monitoring systems' inability to systematically identify anomalies in real-time operational data and trace the root cause of errors without power interruption.

Description

Technical Field

[0001] This application relates to the field of power metering error correction, and in particular to a method and system for intelligent compensation of power metering box errors based on time-series anomaly detection. Background Technology

[0002] As the core equipment for metering electricity to end users in the power system, the accuracy of electricity metering directly affects the fairness of electricity trading and settlement and the economic benefits of power supply companies, making it a crucial link in power operation and management. Currently, the power industry generally adopts a management model combining periodic on-site verification and remote automated meter reading data comparison for monitoring and maintaining the operating status of electricity metering boxes. This includes maintenance personnel carrying standard sources and comparison testing instruments to periodically conduct basic error tests on the metering boxes on-site, and integrating communication modules within the metering boxes to upload operating parameters such as voltage, current, and power to the main station system for simple threshold over-limit judgment and statistical trend analysis. However, this traditional monitoring and maintenance method is limited by fixed on-site verification cycles and coarse-grained threshold judgment logic, making it difficult to capture gradual metering deviations caused by complex operating conditions such as drastic load fluctuations, sudden changes in ambient temperature, or slow degradation of electronic component performance. Furthermore, on-site verification operations require power outages, which not only affect users' normal electricity consumption but also create a sharp contradiction between verification frequency and operation and maintenance costs. Many metering boxes remain in a blind spot of ineffective monitoring for extended periods during the intervals between verifications.

[0003] Currently, in related technologies, the metering error monitoring of electricity metering boxes has the technical problem of being unable to systematically identify anomalies in real-time operating data and trace the root cause of errors without power interruption. Summary of the Invention

[0004] This application provides an intelligent error compensation method and system for electricity metering boxes based on time-series anomaly detection. It acquires multi-dimensional time-series data and performs segmented anomaly identification to filter initial anomaly segments. A dual-path parallel hardware verification circuit is configured to cross-verify the initial anomaly segments to extract the true anomaly segments. A joint search mechanism is introduced to perform root cause localization and multi-scale dynamic error compensation for the true anomaly segments. These technical means solve the technical problem of existing electricity metering box error monitoring systems being unable to systematically identify anomalies in real-time operating data and trace the root cause of errors without power outages. This achieves the technical effect of systematically identifying anomalies in real-time multi-dimensional operating data and accurately locating the root cause of error propagation without power outage intervention, thereby improving the targeting and reliability of metering error compensation.

[0005] This application provides an intelligent error compensation method for electricity metering boxes based on time-series anomaly detection, comprising: acquiring a multi-dimensional time-series operational data stream through a built-in acquisition terminal of the target electricity metering box; performing segmented anomaly identification on the multi-dimensional time-series operational data stream in conjunction with historical operational data of the electricity metering box, and filtering initial abnormal operational data segments; configuring a hardware verification logic circuit inside the target electricity metering box, wherein the hardware verification logic circuit is a dual-channel parallel signal comparison channel, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of adjacent metering boxes in the same area; using the hardware verification logic circuit to perform cross-verification on the initial abnormal operational data segments to obtain the true abnormal operational data segments; introducing a joint search mechanism to perform metering error root cause search on the true abnormal operational data segments, determining the target metering error root cause, and performing multi-scale dynamic error compensation based on the target metering error root cause.

[0006] In a possible implementation, the multi-dimensional time-series operational data stream is segmented and anomaly identified by combining historical operational data from the electricity metering box. Initial abnormal operational data segments are then selected, and the following processing is performed: A standardized data program is constructed based on the electricity metering box data application standard. This standardized data program includes time-series dimension alignment, missing data completion, and pulse anomaly point removal. The multi-dimensional time-series operational data stream is then standardized according to the standardized data program to obtain a standard multi-dimensional time-series operational data stream. A preset dynamic sliding window is used to adaptively segment the standard multi-dimensional time-series operational data stream to obtain a set of multi-dimensional time-series operational data segments. Anomalies are identified in the set of multi-dimensional time-series operational data segments by combining historical operational data from the electricity metering box, and initial abnormal operational data segments are selected.

[0007] In a possible implementation, anomaly identification is performed on the multidimensional time-series operational data segment set by combining historical operational data from the electricity metering box, and initial abnormal operational data segments are screened. The following processes are then performed: extracting a multidimensional associated operational feature set from the multidimensional time-series operational data segment set; calculating a set of segmented data autocorrelation coefficients for the same time period as the multidimensional time-series operational data segment set by combining historical operational data from the electricity metering box; performing dynamic feature extraction on the multidimensional time-series operational data segment set based on the set of segmented data autocorrelation coefficients to obtain a dynamic baseline operational feature set; and performing recursive feature elimination and anomaly morphology identification on the multidimensional associated operational feature set and the dynamic baseline operational feature set to screen initial abnormal operational data segments.

[0008] In a possible implementation, the hardware verification logic circuit is used to cross-verify the initial abnormal operation data segment to obtain the real abnormal operation data segment, and the following processing is performed: the hardware verification logic circuit calls the original sampling signal of the main circuit and the synchronization reference signal of the transformer area corresponding to the initial abnormal operation data segment; a point-by-point cross-comparison is performed on the original sampling signal of the main circuit and the synchronization reference signal of the transformer area to obtain the deviation comparison verification result; if the deviation comparison verification result is within a preset continuous deviation range, it is a false interference anomaly, and the initial abnormal operation data segment is removed and corrected; if the deviation comparison verification result exceeds the preset continuous deviation range, it is a real metering anomaly, and the initial abnormal operation data segment is marked and output to obtain the real abnormal operation data segment.

[0009] In a possible implementation, a joint search mechanism is introduced to perform a root cause search for metering errors on the actual abnormal operation data segment, determine the target root cause of metering errors, and perform the following processing: perform anomaly topology analysis based on the metering anomaly case library of the target energy metering box, and construct an anomaly link topology diagram of the energy metering box; starting from the time window of the actual abnormal operation data segment, traverse the historical operation data of adjacent time periods to determine metering anomalies and obtain directional metering anomaly type features; introduce a joint search mechanism to perform a root cause search for metering errors within the anomaly link topology diagram of the energy metering box based on the directional metering anomaly type features, and determine the target root cause of metering errors.

[0010] In a possible implementation, an anomaly topology analysis is performed based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology graph of the energy metering box. The following processes are performed: the metering anomaly case library of the target energy metering box is decomposed into a labeled form to obtain a structured metering anomaly dataset; a topology node hierarchical architecture is built based on the actual hardware level of the target energy metering box; error-related edge weights are assigned to the topology node hierarchical architecture according to the structured metering anomaly dataset to generate a basic anomaly link topology graph; the basic anomaly link topology graph is verified and iteratively optimized to construct the energy metering box anomaly link topology graph.

[0011] In a possible implementation, a joint search mechanism is introduced to perform a root cause search for metering errors within the abnormal link topology of the energy metering box based on the directional metering anomaly type characteristics. This determines the target root cause of the metering error and involves the following steps: Node matching and activation are performed on the abnormal link topology of the energy metering box according to the directional metering anomaly type characteristics to obtain a queue of nodes to be visited; a joint search mechanism is introduced to traverse the queue of nodes to be visited and perform a breadth-first search for initial horizontal screening to obtain a set of candidate search nodes; based on the set of candidate search nodes, a depth-first search for error roots is performed within the abnormal link topology of the energy metering box to determine the target root cause of the metering error.

[0012] In a possible implementation, a depth-first search for the root cause of the metering error is performed within the abnormal link topology of the electricity metering box based on the candidate search node set to determine the target metering error root cause. The following processing is then performed: starting from the candidate search node set, a depth-first traversal search is performed layer by layer upwards within the abnormal link topology of the electricity metering box to obtain a set of associated parent nodes; the associated parent node set is anomaly determined, and if the associated parent node set is anomaly node, a depth-first search is continued along the branches of the associated parent node set to generate a feasible metering error root cause propagation chain set; the feasible metering error root cause propagation chain set is subjected to credibility verification and screening to determine the target metering error root cause.

[0013] In a possible implementation, multi-scale dynamic error compensation is performed based on the target metering error root cause, and the following processing is performed: a metering error compensation strategy library is constructed, and scenario type matching is performed between the target metering error root cause and the metering error compensation strategy library to determine the multi-scale metering error compensation strategy; dynamic error closed-loop compensation is performed on the sampling data of the target energy metering box based on the multi-scale metering error compensation strategy.

[0014] This application also provides an intelligent error compensation system for electricity metering boxes based on time-series anomaly detection, comprising: a segmented anomaly identification module, used to acquire multi-dimensional time-series operating data streams through the built-in acquisition terminal of the target electricity metering box, and to perform segmented anomaly identification on the multi-dimensional time-series operating data streams in combination with historical operating data of the electricity metering box, thereby filtering initial abnormal operating data segments; a hardware verification logic circuit configuration module, used to configure hardware verification logic circuits inside the target electricity metering box, wherein the hardware verification logic circuits are dual-channel parallel signal comparison channels, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of adjacent metering boxes in the same area; a cross-verification module, used to perform cross-verification on the initial abnormal operating data segments using the hardware verification logic circuits to obtain the true abnormal operating data segments; and a multi-scale dynamic error compensation module, used to introduce a joint search mechanism to perform metering error root cause search on the true abnormal operating data segments, determine the target metering error root cause, and perform multi-scale dynamic error compensation based on the target metering error root cause.

[0015] The proposed intelligent error compensation method and system for electricity metering boxes based on time-series anomaly detection, as outlined in this application, firstly acquires a multi-dimensional time-series operational data stream through a built-in acquisition terminal in the target electricity metering box. This data stream is then segmented and anomalies are identified by combining it with historical operational data from the metering box, filtering out initial anomalous operational data segments. Next, a hardware verification logic circuit is configured inside the target electricity metering box. This hardware verification logic circuit consists of dual parallel signal comparison channels, with the first channel connected to the sampling signal of the main metering circuit and the second channel connected to the synchronization reference signal of adjacent metering boxes in the same area. The initial anomalous operational data segments are then cross-verified using the hardware verification logic circuit to obtain the true anomalous operational data segments. Finally, a joint search mechanism is introduced to perform a root cause search for metering errors on the true anomalous operational data segments, determining the target metering error root cause, and performing multi-scale dynamic error compensation based on the target metering error root cause. Through the above process, the method and system proposed in this application achieve the technical effect of systematically identifying anomalies in real-time collected multi-dimensional operational data and accurately locating the root cause of error propagation without the need for power outage intervention, thereby improving the pertinence and reliability of metering error compensation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the intelligent compensation method for power metering box errors based on timing anomaly detection provided in this application embodiment.

[0018] Figure 2 This is a schematic diagram of the measurement deviation determination comparison line provided in the embodiments of this application.

[0019] Figure 3 This is a schematic diagram of the abnormal link topology of the power metering box provided in the embodiments of this application.

[0020] Figure 4 This is a schematic diagram of the intelligent compensation system for power metering box errors based on timing anomaly detection provided in an embodiment of this application.

[0021] Figure labeling: Segmented anomaly identification module 10, hardware verification logic circuit configuration module 20, cross-verification module 30, multi-scale dynamic error compensation module 40. Detailed Implementation

[0022] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0023] This application provides an intelligent compensation method for errors in electricity metering boxes based on timing anomaly detection, such as... Figure 1 As shown, the method includes: Step S100: Acquire multi-dimensional time-series operation data stream through the built-in acquisition terminal of the target power metering box, and perform segmented anomaly identification on the multi-dimensional time-series operation data stream in combination with the historical operation data of the power metering box to filter the initial abnormal operation data segment.

[0024] Specifically, a multi-dimensional time-series operational data stream is collected in real time by a data acquisition terminal deployed inside the target energy metering box. This acquisition terminal is a data acquisition unit integrated within the energy metering box, with a sampling frequency of no less than 50 times per second. The dimensions collected include at least the RMS voltage value, RMS current value, power factor, active power, reactive power, frequency, and zero-sequence current. The acquisition terminal arranges the above multi-dimensional sampled values ​​in the order of sampling time to form a multi-dimensional time-series operational data stream, and transmits it in real time to the edge computing unit built into the energy metering box via an internal communication bus, serving as the input data source for subsequent processing. Upon receiving the multi-dimensional time-series operational data stream, the edge computing unit synchronously retrieves the historical operational data of the target energy metering box from its local storage medium. This historical operational data consists of standardized multi-dimensional time-series data collected and archived within the same area and the same energy metering box during the same time period over the past 30 calendar days. The edge computing unit uses this historical operating data as a reference baseline to perform segmented anomaly identification on a dimension-by-dimensional basis on the currently acquired multidimensional time-series operating data stream. The continuous multidimensional time-series operating data stream is divided into several data segments according to time sequence. The deviation of the statistical characteristics of each dimension within each data segment from the statistical characteristics of the corresponding data segments in the historical reference baseline is calculated. Data segments with deviations exceeding a preset threshold are selected and marked as initial abnormal operating data segments. The preset threshold is based on three times the standard deviation of the statistical deviation under normal operating conditions of the energy metering box. The specific value is calculated based on the statistical distribution of data segments collected during the historical fault-free operation period of the energy metering box.

[0025] In one possible implementation, the multi-dimensional time-series operational data stream is segmented and anomaly identified by combining historical operational data from the energy metering box. Initial abnormal operational data segments are then filtered. Step S100 further includes step S110, which constructs a standardized data program based on the energy metering box data application standard. The standardized data program includes time-series dimension alignment, missing data completion, and pulse anomaly point removal. Specifically, the edge computing unit first reads a pre-recorded energy metering box data application standard file from its local memory. This file follows the data format and quality requirements in the State Grid Corporation of China's enterprise standard, "Technical Specifications for Energy Metering Boxes." The edge computing unit parses the mandatory provisions in the standard file regarding data sampling frequency, data reporting interval, data valid value range, data accuracy level, and timestamp format, converting these provisions into specific data processing parameter thresholds. Based on this, the edge computing unit constructs a standardized data program. This program resides in the operating system kernel of the edge computing unit as a software module, consisting of three sequentially executed sub-modules. The first submodule is the time-series dimension alignment module. Its execution logic is as follows: according to the unified sampling time benchmark specified in the standard, that is, every whole second corresponds to the start time of the current sampling period, the data of each dimension in the original multi-dimensional time-series running data stream sent by the acquisition terminal is remapped to a unified time grid according to its own timestamp. For data points whose timestamp deviates from the unified grid time within ±1ms, they are directly assigned to the nearest grid point. Data points exceeding this deviation range are supplemented by linear interpolation between adjacent points. The second submodule is the missing data completion module. Its execution logic is as follows: iterate through each preset sampling time in the time-series dimension aligned data stream, and check whether there is a dimension missing or a whole frame data missing at that time. If a missing data occurs, cubic spline interpolation is used. The corresponding dimension data of the five normal sampling times before and after the missing time are used as interpolation nodes. A cubic spline interpolation function is constructed to calculate the completion value of the missing time and fill it into the corresponding position in the data stream. The third submodule is the pulse anomaly removal module. Its execution logic is as follows: For the data stream after missing data completion, calculate the first-order difference value of each sampling point dimension by dimension. If the difference value of a sampling point exceeds five times the standard deviation of the historical difference value mean for that dimension, then that point is determined to be a pulse anomaly. This point is removed from the data stream and replaced with the median value of the three normal points before and after it. The standardized data program encapsulates the three submodules into a serial processing pipeline. The data stream is processed sequentially by the above three modules and then outputs the standardized result. By constructing a standardized data program, the inconsistency in data format caused by hardware differences between different acquisition terminals is resolved, providing standardized, continuous, and complete data input.

[0026] Step S120: The multi-dimensional time-series running data stream is standardized according to the standardized data program to obtain a standard multi-dimensional time-series running data stream. Specifically, the edge computing unit calls the standardized data program constructed in step S110, and passes the multi-dimensional time-series running data stream sent in real time by the acquisition terminal in step S100 as input parameters to the program. The standardized data program first starts the time-series dimension alignment module, reads each frame of data in the multi-dimensional time-series running data stream, extracts the acquisition timestamp and dimension identifier carried by each frame of data, and rearranges the data values ​​of each dimension according to a unified time grid. After alignment, the standardized data program transfers the processed intermediate data stream to the missing data completion module. The missing data completion module scans each time point on the time grid one by one to check whether there are missing dimensions or frames, and completes the missing values ​​according to the interpolation method in step S110. After completion, the standardized data program transfers the data stream to the pulse anomaly removal module, which scans the data values ​​dimension by dimension for differential detection and anomaly replacement processing. After the three modules have been executed sequentially, the standardized data program outputs all processed data in ascending chronological order as a standard multidimensional time-series running data stream, which is then stored in the random access memory of the edge computing unit. Each sampling point in this standard multidimensional time-series running data stream contains a complete set of data dimensions, and each dimension is evenly distributed on the time axis. The time interval between adjacent sampling points is equal to the standard sampling interval of the acquisition terminal.

[0027] Step S130: A preset dynamic sliding window is used to adaptively segment the standard multi-dimensional time-series running data stream to obtain a set of multi-dimensional time-series running data segments. Specifically, the edge computing unit reads the standard multi-dimensional time-series running data stream output in step S120 from the random access memory. The edge computing unit presets a dynamic sliding window to perform adaptive time-series segmentation. This dynamic sliding window has a variable window length, and its initial window length is set to 100 times the standard sampling interval, corresponding to a time span of 100 sampling points. The edge computing unit places this window at the beginning of the standard multi-dimensional time-series running data stream, covering continuous data points starting from the first sampling point. Then, the edge computing unit calculates the statistical characteristic values ​​of the data within the window, including the mean, variance, and peak-to-peak value of each dimension, and compares these statistical characteristic values ​​with the corresponding statistical characteristic values ​​of the top 10% of the data in the entire data stream to calculate the difference. The difference is equal to the weighted sum of the variance differences of each dimension divided by the total number of dimensions, where the weighting coefficient is the preset weight value of each dimension in the error sensitivity of the power metering box. If the calculated difference is less than a preset difference threshold of 5%, the edge computing unit extends the right boundary of the window by 5 sampling points to the right, recalculates the difference, and repeats this process until the difference first exceeds 5% or the right boundary of the window reaches the end of the data stream. When the difference first exceeds 5%, the data segment between the left boundary and the previous sampling point of the right boundary of the current window is cut out as an independent multidimensional time-series running data segment. Then, the edge computing unit moves the left boundary of the window to the position of the right boundary of the current window and resets the window length to the initial window length, repeating the above segmentation process until the entire standard multidimensional time-series running data stream is completely segmented. After segmentation, the edge computing unit puts all the segmented multidimensional time-series running data segments into a set according to their chronological order, forming a multidimensional time-series running data segment set. Each data segment in this set is attached with a start timestamp and an end timestamp to uniquely identify its position in the original data stream. Adaptive time-series segmentation is used to ensure that the statistical characteristics within each data segment are relatively stable after segmentation, avoiding the mixing of normal and abnormal operating conditions within a data segment, thereby improving the resolution of anomaly identification.

[0028] Step S140 involves identifying anomalies in the multi-dimensional time-series operational data segment set by combining historical operational data from the electricity metering box, and filtering initial abnormal operational data segments. Specifically, the edge computing unit obtains the multi-dimensional time-series operational data segment set output in step S130 and the historical operational data of the electricity metering box retrieved from the local storage medium. The edge computing unit searches for historical concurrent data segments corresponding to the timestamps of each data segment in the historical operational data, defined as standardized data segments corresponding to the same time interval as the current data segment on each day within the past 30 natural days. For each current data segment in the multi-dimensional time-series operational data segment set, the edge computing unit subtracts the statistical feature vector of each dimension from the mean of the statistical feature vectors of the same dimension of all corresponding historical concurrent data segments to obtain a feature deviation vector. Then, the components of this feature deviation vector are weighted and summed according to their dimension weights to obtain the comprehensive anomaly score for that data segment. The edge computing unit marks current data segments whose overall anomaly score exceeds a preset anomaly score threshold by 3% as initial anomalies, and outputs all marked anomaly data segments into the initial anomaly running data segment set. The preset anomaly score threshold is determined by the 90th percentile of the overall anomaly score of all data segments during a period without historical anomaly records for the energy metering box. By using historical normal data from the same energy metering box at the same time point as a self-reference benchmark, interference caused by normal operating condition changes such as load fluctuations and grid voltage fluctuations on anomaly identification is eliminated, making the selected initial anomaly data segments closer to actual metering system anomalies rather than changes in the electricity usage environment.

[0029] In one possible implementation, anomaly identification is performed on the multidimensional time-series operation data segment set by combining historical operation data of the power metering box, and initial abnormal operation data segments are screened. Step S140 further includes step S141, extracting a multidimensional correlation operation feature set of the multidimensional time-series operation data segment set. Specifically, the edge computing unit performs multidimensional correlation operation feature extraction on each data segment in the multidimensional time-series operation data segment set output in step S130. The extraction process is as follows: for each data segment, the edge computing unit first calculates the basic statistics of each dimension within the data segment, including the mean, variance, skewness, kurtosis, maximum value, minimum value, range, and root mean square value, a total of eight statistics. Then, the edge computing unit calculates the cross-correlation quantities between different dimensions within the data segment, including the correlation coefficient between the effective voltage value and the effective current value, the correlation coefficient between active power and reactive power, the correlation coefficient between power factor and load rate, and the correlation coefficient between zero-sequence current and three-phase current imbalance, a total of four correlation quantities. The eight basic statistics and four correlation quantities mentioned above are concatenated into a feature vector according to a preset dimension order. This feature vector is the multidimensional correlation operation feature of the data segment. The edge computing unit performs the above extraction operation on each data segment in the multidimensional time-series running data segment set to obtain the feature vector corresponding to each data segment. These feature vectors are then collected into a multidimensional correlation operation feature set according to the time sequence of the data segments. Each feature vector in this set carries the timestamp identifier of its source data segment.

[0030] Step S142 involves calculating the set of segmented data autocorrelation coefficients for the same time period as the multidimensional time-series operational data segment set, based on the historical operational data of the electricity metering box. Specifically, the edge computing unit reads the historical operational data of the electricity metering box from its local memory. This historical operational data is organized as standardized multidimensional time-series data for each of the past 30 natural days, and the data for each day has been pre-segmented into historical data segments according to the segmentation rules adopted in step S130. The edge computing unit traverses each current data segment in the multidimensional time-series operational data segment set. For each current data segment, it searches for all historical data segments with the same timestamp interval from the historical operational data, resulting in 30 historical data segments. For each historical data segment, the edge computing unit calculates the mean vector of each dimension within that historical data segment. Then, the edge computing unit performs a dot product operation between the mean vector of the current data segment and the mean vector of each historical data segment, and divides the dot product result by the product of the modulus of the mean vector of the current data segment and the modulus of the mean vector of the historical data segment to obtain the similarity coefficient between the historical data segment and the current data segment. The edge computing unit arranges the 30 similarity coefficients into a sequence of length 30, arranged chronologically from oldest to newest. It then calculates the first-order autocorrelation coefficient of this sequence, which is the Pearson correlation coefficient between two adjacent similarity coefficients. This first-order autocorrelation coefficient is used as the segmented data autocorrelation coefficient for the current data segment. The edge computing unit performs this calculation for each current data segment in the multi-dimensional time-series running data segment set, obtaining the segmented data autocorrelation coefficient for each segment. All segmented data autocorrelation coefficients are then integrated into a set of segmented data autocorrelation coefficients.

[0031] Step S143: Based on the set of autocorrelation coefficients of the segmented data, perform dynamic feature extraction on the set of multidimensional time-series running data segments to obtain a dynamic baseline running feature set. Specifically, the edge computing unit obtains the set of autocorrelation coefficients of the segmented data output in step S142 and the set of multidimensional correlation running features output in step S141. The edge computing unit performs dynamic feature extraction on each data segment in the set of multidimensional time-series running data segments. The extraction process is as follows: For the i-th data segment, the edge computing unit takes its corresponding segmented data autocorrelation coefficient value, as well as the segmented data autocorrelation coefficient values ​​corresponding to its two temporally adjacent data segments and the two temporally adjacent data segments, for a total of five values. These five values ​​are combined into an autocorrelation window vector of length five. Then, the edge computing unit calculates the weighted moving average of the autocorrelation window vector, with the weighting coefficients being 0.1 for the two previous times, 0.2 for the previous time, 0.4 for the current time, 0.2 for the next time, and 0.1 for the two subsequent times. The edge computing unit multiplies the calculated weighted moving average by each component of the multidimensional correlation operating feature vector of the data segment to obtain the dynamic baseline operating feature vector of the data segment. This dynamic baseline operating feature vector reflects the expected characteristic amplitude of the data segment under the current autocorrelation fluctuation level. The edge computing unit performs the above calculation on each data segment in the multidimensional time-series operating data segment set to obtain the dynamic baseline operating feature vector corresponding to each data segment, and sets these feature vectors into a dynamic baseline operating feature set. By incorporating the autocorrelation change trend of historical data segments into the current feature extraction process, the baseline features can adaptively adjust to changes in grid background noise and load patterns, avoiding misjudgments caused by static baselines when the environment changes.

[0032] Step S144 involves recursive feature elimination and anomaly identification on the multidimensional associated operational feature set and the dynamic baseline operational feature set to filter initial abnormal operational data segments. Specifically, the edge computing unit performs component-wise difference operations on each feature vector in the multidimensional associated operational feature set obtained in step S141 and the dynamic baseline operational feature vector corresponding to the same data segment in the dynamic baseline operational feature set obtained in step S143 to obtain the residual feature vector for each data segment. The dimension of the residual feature vector is equal to the dimension of the multidimensional associated operational feature vector. The edge computing unit arranges the residual feature vectors of all data segments into a residual feature matrix in chronological order. Then, the edge computing unit performs a recursive feature elimination operation on the residual feature matrix: first, it calculates the variance of each dimension in the residual feature matrix, removes the dimension with the smallest variance from the matrix, then recalculates the mutual information value between each dimension and the pre-labeled historical anomaly label for the remaining dimensions, removes the dimension with the smallest mutual information value again, and repeats this process until the number of remaining dimensions reaches the preset number of retained dimensions, which is 60% of the original number of dimensions. The edge computing unit records the final retained dimension indices and extracts the corresponding residual feature sub-vectors from the original residual feature vectors according to these indices. Next, the edge computing unit uses a density-based anomaly detection algorithm, namely the local outlier factor algorithm, to identify anomalies in all residual feature sub-vectors. In the local outlier factor algorithm, for each residual feature sub-vector, the local reachability density of its reachability distance to the K nearest neighboring sample points is calculated, where K is taken as 5% of the total number of data segments, rounded up. Then, the local reachability density of that point is compared with the average local reachability density of all points in its neighborhood to obtain the local outlier factor value. The edge computing unit marks the data segments corresponding to residual feature sub-vectors with local outlier factor values ​​greater than a preset threshold of 1.5 as anomalies and collects these data segments as the initial abnormal running data segment output. Recursive feature elimination reduces noise interference from redundant features in anomaly detection, and density-based anomaly detection identifies anomalous data segments that deviate from the normal clusters in the feature space, making the selected initial abnormal data segments statistically significant outliers.

[0033] Step S200: Configure a hardware verification logic circuit inside the target energy metering box. The hardware verification logic circuit is a dual-channel parallel signal comparison channel, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of the adjacent metering box in the same area.

[0034] Specifically, a hardware verification logic circuit is configured on the circuit board level inside the target energy metering box. This hardware verification logic circuit uses a field-programmable gate array (FPGA) as its core and internally embeds dual parallel signal comparison channels. In practice, a mounting position for the FPGA is reserved on the printed circuit board of the target energy metering box, and its power pins, ground pins, and configuration pins are electrically connected to the internal power supply system and configuration circuit of the metering box. After configuration, the FPGA forms two independent signal processing links, each with its own independent analog-to-digital conversion front-end, digital filtering unit, and buffer register. The two links share the same high-precision clock source to ensure the synchronization of signal sampling. The signal input of the first signal comparison channel is connected to the secondary output of the current transformer and voltage divider of the main metering circuit of the target energy metering box via shielded twisted-pair cable. This is used to input the original sampling signals of the main metering circuit, including the secondary current signal induced by the current transformer and the secondary voltage signal output by the voltage divider. The signal input of the second signal comparison channel is connected via a shielded twisted-pair cable to the synchronization reference signal output interface of another adjacent energy metering box located in the same substation area. This synchronization reference signal consists of the raw current and voltage sampling signals collected simultaneously by the main metering circuit inside the adjacent energy metering box. Driven by its internal reference signal output interface circuit, the signals are transmitted to the hardware verification logic circuit of the target energy metering box. After entering the field-programmable gate array (FPGA), both signals undergo analog-to-digital conversion, digital filtering, and clock synchronization latching processing on their respective links, converting them into digital signal sequences that can be compared point-by-point. Once the hardware verification logic circuit is configured, both parallel signal comparison channels are in a ready-to-trigger state, awaiting verification commands from the edge computing unit to initiate the comparison process. By constructing physical signal comparison channels at the hardware level, independent of the software processing flow, common-mode interference and systematic biases that may be introduced by software algorithms are avoided, providing a highly reliable physical basis for cross-verification.

[0035] Step S300: The initial abnormal running data segment is cross-validated using the hardware verification logic circuit to obtain the real abnormal running data segment.

[0036] Specifically, after the edge computing unit filters out the initial abnormal running data segment in step S100, it sends a cross-verification instruction to the hardware verification logic circuit configured in step S200. This instruction carries the timestamp interval corresponding to the initial abnormal running data segment. Upon receiving the instruction, the field-programmable gate array (FPGA) in the hardware verification logic circuit extracts the digital signal sequences stored in the first and second signal comparison channels within the timestamp interval from its internally cached historical signal records, based on the timestamp interval in the instruction. Subsequently, the two signal comparison channels within the FPGA simultaneously initiate point-by-point cross-verification operations, calculating the difference between the first and second digital signal values ​​at the same sampling time point to obtain the instantaneous deviation value at each sampling time point. The hardware verification logic circuit continuously arranges the instantaneous deviation values ​​at each sampling time point to form a deviation comparison verification result sequence, and transmits this sequence back to the edge computing unit in real time. After receiving the deviation comparison and verification result sequence, the edge computing unit performs a continuity determination on the sequence. If the number of sampling points in the sequence that continuously exceed the preset deviation threshold is less than the preset continuous deviation range threshold, the deviation is determined to be a random false interference anomaly. The edge computing unit removes this initial abnormal running data segment from the anomaly list and marks the data segment at the removal position as interference noise before performing data correction using linear interpolation. If the number of sampling points in the sequence that continuously exceed the preset deviation threshold is greater than or equal to the preset continuous deviation range threshold, the deviation is determined to be a real metering anomaly generated by the metering main circuit of the target energy metering box itself. The edge computing unit marks this initial abnormal running data segment as a real abnormal running data segment and outputs it. The preset deviation threshold is set to 2% of the rated signal amplitude, and the preset continuous deviation range threshold is set to 50 consecutive sampling points. By utilizing the high-speed parallel processing capability and physical signal independence of the hardware circuit, false positive anomalies caused by external electromagnetic interference or communication jitter in the software anomaly identification results are removed, retaining the real metering anomaly data segments that can be verified at the physical level, thus improving the reliability of the anomaly detection results.

[0037] In one possible implementation, the hardware verification logic circuit is used to cross-verify the initial abnormal operation data segment to obtain the real abnormal operation data segment. Step S300 further includes step S310, in which the hardware verification logic circuit calls the main circuit original sampling signal and the transformer area synchronization reference signal corresponding to the initial abnormal operation data segment. Specifically, the edge computing unit uses the start timestamp and end timestamp of the initial abnormal operation data segment selected in step S144 as calling parameters and sends them to the hardware verification logic circuit configured in step S200 via the internal bus. After receiving the calling parameters, the field-programmable gate array device in the hardware verification logic circuit reads the digital signal sequence stored in the timestamp interval from the buffer register of the first signal comparison channel integrated inside it as the main circuit original sampling signal. This signal sequence is the digital waveform data obtained by the analog-to-digital converter sampling the secondary side output signals of the metering main circuit current transformer and voltage divider. Simultaneously, the field-programmable gate array (FPGA) reads the digital signal sequence stored within the same timestamp interval from the buffer register of its internally integrated second signal comparison channel as the area synchronization reference signal. This signal sequence is the digitized waveform data obtained by the analog-to-digital converter sampling the synchronization reference signal transmitted from adjacent energy metering boxes in the same area via shielded twisted-pair cables. Both signal sequences are arranged in ascending order of time, indexed by the sampling time, and share the same sampling trigger pulse generated by the same clock source, thus ensuring a one-to-one correspondence between the two signal sequences on the time axis.

[0038] Step S320: Perform point-by-point cross-comparison on the original sampling signal of the main circuit and the synchronization reference signal of the transformer area to obtain the deviation comparison verification result. Specifically, the dual parallel signal comparison channels configured inside the field-programmable gate array (FPGA) in the hardware verification logic circuit simultaneously start the point-by-point cross-comparison operation. For each sampling time point within the timestamp interval, the first signal comparison channel outputs the digital value of the original sampling signal of the main circuit at that time point to a comparator input terminal inside the FPGA, and the second signal comparison channel outputs the digital value of the transformer area synchronization reference signal at the same time point to the other input terminal of the same comparator. The comparator calculates the absolute value of the difference between the two input values ​​as the instantaneous deviation value at that sampling time point. The comparator processes each sampling point sequentially according to the sampling time order, and continuously outputs the instantaneous deviation value of each sampling time point to form a deviation comparison verification result sequence. The length of this result sequence is equal to the total number of sampling points contained in the timestamp interval, and each element in the sequence corresponds to a unique sampling time point. The field-programmable gate array (FPGA) temporarily stores the deviation comparison and verification result sequence in its internal output buffer and transmits it back to the edge computing unit via the data bus.

[0039] Step S330: If the deviation comparison verification result is within a preset continuous deviation range, it is considered a false interference anomaly, and the initial abnormal running data segment is removed and corrected. Specifically, the edge computing unit receives the deviation comparison verification result sequence from the hardware verification logic circuit and performs a continuity determination on the sequence. The edge computing unit scans the deviation comparison verification result sequence point by point. When the instantaneous deviation value of a certain sampling point is less than or equal to a preset deviation threshold, it continues scanning and simultaneously starts a continuous counting variable, resets the variable to zero, and then restarts the counting. This continuous counting variable is named the deviation continuous count and is used to accumulate the number of currently consecutive sampling points that do not exceed the tolerance. When the value of the deviation continuous count reaches the preset continuous deviation range threshold, the edge computing unit determines that the initial abnormal running data segment corresponding to all sampling points in the current scanning interval whose instantaneous deviation values ​​do not exceed the preset deviation threshold belongs to a false interference anomaly. The edge computing unit removes the initial abnormal data segment from the anomaly list and performs a correction operation: using the mean of the 10 normal sampling points before the start time and the mean of the 10 normal sampling points after the end time of the data segment as endpoints, linear interpolation is used to generate filler data. This filler data is then written to the corresponding position of the original data segment in the standard multidimensional time-series data stream, overwriting the original data. False anomalies confirmed by hardware comparison as common-mode interference or communication jitter are excluded from subsequent processing to avoid false alarms.

[0040] Step S340: If the deviation comparison verification result exceeds the preset continuous deviation range, it is considered a true measurement anomaly. The initial abnormal running data segment is then marked and output to obtain a true abnormal running data segment. Specifically, following the scanning and counting process in step S330, when the edge computing unit scans the deviation comparison verification result sequence, if it detects that the instantaneous deviation value of a certain sampling point is greater than the preset deviation threshold, it resets the deviation continuous count variable to zero and starts another out-of-tolerance continuous count variable, named the out-of-tolerance continuous count, to accumulate the number of currently consecutive out-of-tolerance sampling points. When the value of the out-of-tolerance continuous count reaches the preset continuous deviation range threshold, the edge computing unit determines that the initial abnormal running data segment is a true measurement anomaly. The edge computing unit does not perform data correction on this data segment, but instead writes a "true anomaly" identifier in the label field of this data segment and outputs this data segment and its label as a true abnormal running data segment. If the edge computing unit detects that there are both out-of-tolerance segments with less than 50 consecutive sampling points and out-of-tolerance segments with more than 50 consecutive sampling points in the deviation comparison and verification result sequence, then the out-of-tolerance segments with more than 50 consecutive sampling points will be used as the judgment criteria, and the corresponding initial abnormal running data segments will be marked as real anomalies.

[0041] For example, assume that the timestamp interval corresponding to the initial abnormal operation data segment contains 100 sampling points, numbered P1 to P100. The hardware verification logic circuit calculates the instantaneous deviation between the original sampled signal value of the main circuit and the synchronization reference signal value of the transformer area for each sampling point. The preset deviation threshold is 2% of the rated amplitude, which corresponds to a digital quantity deviation threshold of 50 digital units. The preset continuous deviation range threshold is 50 consecutive sampling points. The instantaneous deviation values ​​of each sampling point are shown in Table 1 and... Figure 2 As shown (only some key sampling points are shown).

[0042] Table 1 ; The edge computing unit scans the deviation sequence, continuously increasing the count of points without deviation starting from P1, reaching 25 at P25. A deviation first occurs at P26, the continuous count of points without deviation resets to zero, and the continuous count of points with deviation starts counting from 1 at P26. Deviations continue from P27 to P75, reaching 50 at P75, which is the preset continuous deviation range threshold. The edge computing unit determines this initial abnormal data segment as a genuine measurement anomaly and marks it as "Genuine Anomaly." If deviations only occur at the two isolated points P26 and P27, while the remaining sampling points are not, the continuous count of points with deviation will not reach 50, and the continuous count of points without deviation will restart counting at P28 until reaching 50 at P77. The edge computing unit determines this situation as a false interference anomaly and performs a correction.

[0043] Step S400: A joint search mechanism is introduced to perform a root cause search for measurement errors on the real abnormal operation data segment, determine the target root cause of measurement errors, and perform multi-scale dynamic error compensation based on the target root cause of measurement errors.

[0044] Specifically, after obtaining the actual abnormal operation data segment output in step S300, the edge computing unit initiates a joint search mechanism. This joint search mechanism is a two-stage graph search algorithm consisting of breadth-first search and depth-first search, running on the processor inside the edge computing unit. The edge computing unit first uses the center time of the time window of the actual abnormal operation data segment as the starting time point for the search, and obtains abnormal case records near that time window from the local metering anomaly case library of the target energy metering box through timestamp index matching. Then, using the anomaly type field in the abnormal case record as the search keyword, it locates the matching starting node in the pre-constructed abnormal link topology graph of the energy metering box. The joint search mechanism enters the first stage: starting from the starting node, it performs a breadth-first search lateral initial screening along the edges of the abnormal link topology graph, traversing all adjacent nodes that are one hop away from the starting node, and recording all adjacent nodes and their connecting edge weights in the candidate search node set. The joint search mechanism then enters the second phase: starting with each node in the candidate search node set, a depth-first traversal search is performed layer by layer upwards along the edges representing the error propagation direction in the abnormal link topology graph. During the search process, the error propagation coefficient between each traversed node and its parent node is accumulated. When the accumulated result converges to a stable value or reaches the preset maximum search depth, the search for that branch is stopped, and the node sequence on the search path is recorded as a feasible measurement error root cause propagation chain. The edge computing unit calculates the credibility of all feasible measurement error root cause propagation chains, sorts them from high to low credibility, and selects the error cause field corresponding to the root cause node on the propagation chain with the highest credibility to determine the target measurement error root cause. After identifying the root cause of the target metering error, the edge computing unit matches the corresponding compensation strategy from a pre-stored metering error compensation strategy library based on the root cause. This compensation strategy includes frequency band compensation coefficients based on frequency response, temperature compensation coefficients based on temperature drift, and nonlinear compensation coefficients based on load. These compensation coefficients are used to perform dynamic closed-loop error compensation on the sampled data of the target energy metering box, point-by-point, according to their respective physical dimensions. Specifically, the compensation coefficients are multiplied and added to the original sampled values ​​to generate compensated metering data, which is then written back to the metering data storage area for subsequent billing. By introducing a graph search mechanism, the propagation path of errors between functional nodes within the energy metering box is systematically explored, establishing a causal relationship between anomalies and underlying causes. Based on the identified root causes, targeted multi-dimensional error compensation is implemented, achieving a closed loop from anomaly detection to error correction.

[0045] In one possible implementation, a joint search mechanism is introduced to perform a root cause search for metering errors on the actual abnormal operation data segment to determine the target metering error root cause. Step S400 further includes step S410, which involves performing anomaly topology analysis based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology diagram of the energy metering box. Specifically, the edge computing unit reads the metering anomaly case library of the target energy metering box from the local storage medium. This case library is a structured database file, where each record contains the anomaly occurrence time, anomaly manifestation type, anomaly root cause, anomaly propagation path description, and a tag indicating whether the anomaly has been confirmed and repaired. The edge computing unit performs anomaly topology analysis on this case library: first, it extracts the anomaly root cause field and anomaly manifestation type field from each case record to construct a root cause-manifestation pair set. Then, it counts the co-occurrence frequency between the same root cause and the same manifestation in all pairs, using the co-occurrence frequency as the weight of the edge. Next, using all different root causes and manifestation types of anomalies as nodes in the topology graph, and root cause-manifestation relationships with a co-occurrence frequency greater than 0 as edges, weight values ​​are assigned to the corresponding edges to generate the initial structure of a directed acyclic graph. The edge computing unit performs connectivity checks and redundant edge merging on this initial structure, ultimately obtaining the anomaly link topology graph of the energy metering box. Each node in this topology graph is assigned a unique Chinese node identifier, and the direction of each directed edge points from the root cause node to the manifestation node, indicating the direction of error propagation from the root cause to the manifestation.

[0046] Step S420: Starting from the time window of the actual abnormal operation data segment, the historical operation data of adjacent time periods are traversed to determine the measurement anomaly and obtain the directional measurement anomaly type feature. Specifically, the edge computing unit obtains the actual abnormal operation data segment output in step S340, reads the start and end timestamps of the data segment, and uses the time window as the starting point. The edge computing unit retrieves the historical operation data segments corresponding to the time window and the two adjacent time windows before and after the time window (each window length is the same as the time window length of the actual abnormal operation data segment) from the local storage medium, for a total of five data segments. For these five historical operation data segments, the edge computing unit extracts their multidimensional associated operation feature vectors respectively, and performs a difference operation with the multidimensional associated operation feature vectors of the corresponding position data segments obtained in step S141 to obtain five difference vectors. The edge computing unit calculates the mean vector and standard deviation vector of these five difference vectors, and then calculates the degree of deviation between the multidimensional associated operation feature vector of the actual abnormal operation data segment and the mean vector, marking the dimension type corresponding to the component with a deviation degree exceeding twice the standard deviation vector as the directional measurement anomaly type feature. For example, if the voltage dimension deviates by more than twice the standard deviation while other dimensions do not, the directional measurement anomaly type feature is "voltage anomaly". This directional measurement anomaly type feature is stored in the form of Chinese tags to guide the direction of root cause search.

[0047] Step S430: A joint search mechanism is introduced to perform a root cause search for metering errors within the abnormal link topology of the energy metering box based on the directional metering anomaly type characteristics, thereby determining the target root cause of the metering error. Specifically, the edge computing unit uses the Chinese tags of the directional metering anomaly type characteristics obtained in step S420 as search keywords to find nodes containing these keywords in the abnormal link topology of the energy metering box constructed in step S410, and uses these nodes as the starting search node. The edge computing unit initiates the joint search mechanism, which is divided into a horizontal preliminary screening stage and a vertical deep search stage. In the horizontal preliminary screening stage, the edge computing unit starts from the starting node and traverses all nodes with a distance of one edge from the starting node along the undirected adjacency relationship of the topology graph, adding these nodes to the candidate search node set. During the vertical depth-first search phase, the edge computing unit searches upwards layer by layer along the edges in the topology graph (from the appearance side to the root cause side) for each node in the candidate search node set. Each time an edge is traversed, the node sequence along the path is recorded in the current propagation chain, until a node with an in-degree of zero (i.e., a node with no further root causes) or a preset maximum search depth of 6 is reached. After the search is complete, the edge computing unit obtains several propagation chains from the starting node to the root cause node. For each propagation chain, the product of the weights of all edges along its path is calculated as the confidence value of that propagation chain. The abnormal root cause field corresponding to the terminal node of the propagation chain with the highest confidence value is determined as the target measurement error root cause.

[0048] In one possible implementation, an anomaly topology analysis is performed based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology diagram of the energy metering box. Step S410 further includes step S411, which involves tagging and decomposing the metering anomaly case library of the target energy metering box to obtain a structured metering anomaly dataset. Specifically, the edge computing unit reads the original file of the metering anomaly case library of the target energy metering box. This original file is an unstructured text log, and each log entry contains the time of anomaly occurrence, a description of the anomaly phenomenon, handling measures, and handling results. The edge computing unit uses a text parsing method based on regular expressions to tag and decompose each log entry. Specifically, the edge computing unit configures multiple sets of regular expression matching templates. The first set of templates matches the timestamp field in the logs and converts it into a unified format for storage. The second set of templates matches keywords in the description of abnormal phenomena, such as "voltage surge," "current imbalance," and "power factor decline," and maps the matched keywords to preset abnormal phenomenon type labels. The third set of templates matches keywords in the handling measures, such as "replace transformer," "tighten terminals," and "upgrade firmware," and maps the matching results to preset abnormal root cause type labels. The fourth set of templates matches keywords in the handling results, such as "repaired" and "not reproduced," and maps the matching results to an abnormal confirmation status label. The edge computing unit applies the above four sets of templates to each log in sequence, extracts the matching results, and combines them into a structured record. This record contains a timestamp field, an abnormal phenomenon type label, an abnormal root cause type label, and an abnormal confirmation status label. After processing all logs, the edge computing unit integrates all structured records into a structured metering abnormality dataset and stores it in a relational database table.

[0049] Step S412: Based on the actual hardware hierarchy of the target energy metering box, a topology node layered architecture is constructed. Specifically, the edge computing unit designs the topology node layered architecture according to the actual hardware hierarchy of the target energy metering box. The hardware hierarchy of the target energy metering box, from top to bottom, consists of a power supply layer, a voltage sampling layer, a current sampling layer, a signal conditioning layer, an analog-to-digital conversion layer, a digital processing layer, a communication interface layer, and a clock reference layer. Each layer is an independent topology layer. Within each topology layer, the edge computing unit establishes topology nodes corresponding to the hardware components of that layer. Each topology node contains a set of feature fields characterizing the normal operating status of the hardware component. The fields in this set include the component's model identifier, nominal accuracy level, calibration date, temperature drift coefficient, and aging coefficient. The connection relationships between the topology layers are determined according to the actual hardware signal flow direction, i.e., the signal flows from the power supply layer into the voltage sampling layer and the current sampling layer, then sequentially through the signal conditioning layer, the analog-to-digital conversion layer, and the digital processing layer, finally reaching the communication interface layer and the clock reference layer. The edge computing unit establishes a tree-like node structure in memory according to the above hierarchy and connection relationship. The root node is located in the power supply layer, and the leaf nodes are located in the communication interface layer and the clock reference layer. This tree-like structure is the topology node hierarchical architecture.

[0050] Step S413: Assign error-related edge weights to the topology node hierarchical architecture according to the structured metrology anomaly dataset to generate a basic anomaly link topology graph. Specifically, the edge computing unit reads each structured record in the structured metrology anomaly dataset output in step S411. For each record, it extracts the anomaly root cause type label and anomaly manifestation type label. In the topology node hierarchical architecture established in step S412, the edge computing unit finds the hardware-level node corresponding to the anomaly root cause type label as the starting node and the hardware-level node corresponding to the anomaly manifestation type label as the ending node. A directed edge is established between the starting node and the ending node, with the direction pointing from the starting node to the ending node. After traversing all structured records, the edge computing unit forms several directed edges in the topology node hierarchical architecture. For each directed edge, the edge computing unit counts the frequency of the edge in the structured metrology anomaly dataset and uses the normalized value of the frequency as the weight value of the edge. The normalization method is to divide the frequency of each edge by the sum of the frequencies of all edges. The edge computing unit combines all directed edges with weight values ​​greater than 0 and their corresponding nodes to generate a basic abnormal link topology graph.

[0051] Step S414 involves verifying and iteratively optimizing the basic abnormal link topology to construct an abnormal link topology for the power metering box. Specifically, the edge computing unit performs verification and iterative optimization operations on the basic abnormal link topology generated in step S413. The verification operation includes a first verification and a second verification. The first verification is a node isolation verification: the edge computing unit scans all nodes in the basic abnormal link topology. If the sum of the out-degree and in-degree of a node is 0, the node is removed from the topology. The second verification is an edge direction consistency verification: the edge computing unit checks whether each directed edge is consistent with the hardware signal flow direction. If the direction of an edge is opposite to the hardware signal flow direction, the direction of the edge is flipped. The iterative optimization operation is based on loop detection: the edge computing unit performs a depth-first traversal search on the verified topology to detect whether there is a directed cycle in the graph. If a directed cycle exists, the average weight of each edge in the cycle is calculated, and edges with weights lower than the average weight are removed from the cycle to break the cycle. The edge computing unit repeatedly performs the above node isolation check, edge direction consistency check, and loop closure detection until the number of nodes and edges in the topology graph no longer changes in two consecutive iterations. At this point, the topology graph converges, and the edge computing unit stores the converged topology graph as the abnormal link topology graph of the power metering box on the local disk.

[0052] In one possible implementation, a joint search mechanism is introduced to perform a root cause search for metering errors within the abnormal link topology of the energy metering box based on the directional metering anomaly type characteristics, determining the target metering error root cause. Step S430 further includes step S431, which involves node matching and activation of the abnormal link topology of the energy metering box according to the directional metering anomaly type characteristics, resulting in a queue of nodes to be accessed for search. Specifically, the edge computing unit obtains the Chinese labels of the directional metering anomaly type characteristics output in step S420, such as "voltage sampling anomaly". The edge computing unit traverses all nodes in the energy metering box abnormal link topology finally output in step S414, comparing the node name field of each node with the Chinese labels of the directional metering anomaly type characteristics. If they match, the node is activated. The edge computing unit records the node identifiers of all activated nodes in the topology and arranges them in the order in which they are traversed, forming a queue structure, which is the queue of nodes to be accessed for search. If the directional metering anomaly type characteristics contain multiple Chinese labels, all matched nodes are added to the queue. The queue is stored in the memory of the edge computing unit as the initial input to the joint search mechanism.

[0053] Step S432: A joint search mechanism is introduced to traverse the queue of nodes to be visited and perform a breadth-first search for preliminary screening to obtain a set of candidate nodes. Specifically, the edge computing unit takes the head node from the queue of nodes to be visited obtained in step S431 as the current starting node. The joint search mechanism enters the preliminary screening stage: the edge computing unit, with the current starting node as the center, searches for all adjacent nodes in the abnormal link topology graph of the power metering box that are directly connected to the current starting node by a directed edge, regardless of the direction of the edge. Whether the edge points from the current starting node to an adjacent node or from an adjacent node to the current starting node, as long as they are directly connected, they are included in the candidate list. The edge computing unit records the node identifiers of all these directly adjacent nodes in the candidate search node list, and also adds the node identifier of the current starting node itself to the list. The edge computing unit takes the next node from the queue of nodes to be visited, repeats the above adjacent node search and recording operation, and appends the adjacent node identifier found this time to the existing record in the candidate search node list. If the adjacent node identifier already exists in the candidate search node list, it will not be added again. After the edge computing unit has traversed all nodes in the queue of search nodes to be accessed, it integrates all node identifiers in the candidate search node list into a candidate search node set. Each node in this set has a direct topological association with at least one node that matches a directional metering anomaly type feature.

[0054] Step S433: Based on the candidate search node set, perform a depth-first search for the root cause of the error within the abnormal link topology of the energy metering box to determine the target metering error root cause. Specifically, the edge computing unit obtains the candidate search node set output in step S432 and uses each node identifier in the set as an independent search starting point. The joint search mechanism enters the vertical depth-first search stage: For each node in the candidate search node set, the edge computing unit performs a depth-first traversal search along the direction of the edge in the abnormal link topology of the energy metering box. The search direction is from the node to its parent node (i.e., the directed edge points to the adjacent node of the node). If a node has multiple parent nodes, the search is performed along the branches of each parent node. During the search process, the edge computing unit maintains a current depth count variable, named the current search depth. The current search depth increases by 1 for each upward traversal. When the current search depth reaches the preset maximum search depth, or when the current node has no parent node, the search of the current branch is stopped, and all nodes passed during the search process are arranged into a node sequence in order from the search starting point to the end node, which is recorded as a feasible metering error root cause propagation chain. After the edge computing unit completes the above search for nodes in all candidate search node sets, it obtains multiple feasible measurement error root cause transmission chains, and selects and determines the target measurement error root cause from these transmission chains.

[0055] In one possible implementation, based on the candidate search node set, a depth-first search for the root cause of the error is performed within the abnormal link topology of the energy metering box to determine the target metering error root cause. Step S433 further includes step S4331, starting from the candidate search node set, performing a depth-first traversal search layer by layer upwards within the abnormal link topology of the energy metering box to obtain a set of associated parent nodes. Specifically, the edge computing unit uses each node in the candidate search node set as the current node and searches for all parent nodes in the abnormal link topology of the energy metering box that point to the current node through directed edges. For each parent node found, the edge computing unit records its node identifier in the associated parent node list. Then, the edge computing unit uses these parent nodes as the new current nodes and continues to search upwards for their parent nodes, appending the newly found parent nodes to the associated parent node list. The edge computing unit repeats the above upward search and append operation until all nodes found at a certain layer have no parent nodes, i.e., reaching the root node layer of the topology. Once all nodes in the candidate search node set have completed the aforementioned layer-by-layer upward traversal, the edge computing unit deduplicates all node identifiers in the associated parent node list to form the associated parent node set. This set contains all upper-level nodes reachable from the candidate search nodes, i.e., all possible error source nodes.

[0056] Step S4332 involves anomaly determination of the associated parent node set. If the associated parent node set is an anomaly node, the search continues layer by layer along the branches of the associated parent node set to generate a feasible metering error root cause propagation chain set. Specifically, the edge computing unit performs anomaly determination on each node in the associated parent node set obtained in step S4331. The anomaly determination method is as follows: the edge computing unit reads the historical anomaly record count of the node in the structured metering anomaly dataset. If the historical anomaly record count is greater than 0, the node is determined to be an anomaly node; if the historical anomaly record count is equal to 0, the node is determined to be a normal node. For nodes determined to be anomalies, the edge computing unit adds the node to the anomaly node queue and uses the node as the new current node to continue searching upwards for its parent node in the electricity metering box anomaly link topology diagram. The newly found parent node is then subjected to anomaly determination again, and this process is repeated until the historical anomaly record count of the root node or the current node is 0. During the aforementioned layer-by-layer search process, the edge computing unit sequentially connects all nodes traversed from the original node in the candidate search node set to the final root node or non-abnormal node, forming a complete directed path. The edge computing unit records each directed path as a feasible measurement error root cause propagation chain, and all feasible measurement error root cause propagation chains together constitute a feasible measurement error root cause propagation chain set.

[0057] Step S4333 involves performing a credibility verification and screening on the feasible measurement error root cause transmission chain set to determine the target measurement error root cause. Specifically, the edge computing unit obtains each transmission chain in the feasible measurement error root cause transmission chain set generated in step S4332. For each transmission chain, the edge computing unit calculates its credibility value, which is equal to the product of the weight values ​​of all directed edges on that transmission chain. The weight values ​​of each directed edge have been assigned and stored in the topology graph in step S413. The edge computing unit assigns the calculated credibility value to the corresponding transmission chain and arranges all transmission chains in descending order of credibility value. The edge computing unit selects the transmission chain with the largest credibility value as the target transmission chain, reads the abnormal root cause type label field corresponding to the terminal node of the target transmission chain, i.e., the top-level root cause node, and determines the content of this field as the target measurement error root cause. If multiple transmission chains have the same and are all the largest, the lengths of these transmission chains are compared, and the target measurement error root cause of the shortest transmission chain is selected as the final output.

[0058] For example, such as Figure 3As shown, assume the abnormal link topology of the electricity metering box includes nodes A to G, where node A is a node in the candidate search node set. The directed edge relationships between the nodes in the topology are as follows: node B points to node A, node C points to node B, node D points to node B, node E points to node C, node F points to node D, and node G points to node E. The edge computing unit executes step S4331 starting from node A, searching upwards to find the parent node B, then searching upwards from B to find the parent nodes C and D, then searching upwards from C to find the parent node E, then searching upwards from D to find the parent node F, then searching upwards from E to find the parent node G, then searching upwards from F to find an empty node, and finally searching upwards from G to find an empty node. The associated parent node set is {B, C, D, E, F, G}. Then, step S4332 is executed, assuming the historical abnormal record counts for each node are: B = 2, C = 0, D = 3, E = 1, F = 0, and G = 0. The edge computing unit determines that B is an abnormal node, C is a normal node, D is an abnormal node, E is an abnormal node, F is a normal node, and G is a normal node. Starting from A, the search continues along the branch B. B's parent node C is a normal node, and the upward search of this branch terminates, with the node sequence ABC, generating the transmission chain ABC. B's other parent node D is an abnormal node, and the search continues along the branch D until F is found to be a normal node, terminating the upward search of this branch, with the node sequence ABDF, generating the transmission chain ABDF. Since C is a normal node, the search stops at C, preventing access to E and G, and no additional transmission chains are generated. Therefore, the feasible set of measurement error root cause transmission chains contains two transmission chains: ABC and ABDF. Step S43333 credibility verification and screening is performed: the product of the weights of all directed edges on the two transmission chain paths is calculated as the credibility value. Transmission chain ABDF has higher credibility, and its end node is a normal node F, which does not possess root cause attributes. The last abnormal node D in the chain is selected, and the abnormal root cause type label corresponding to node D is determined as the target measurement error root cause.

[0059] In one possible implementation, multi-scale dynamic error compensation is performed based on the target measurement error root cause. Step S400 further includes step S440, constructing a measurement error compensation strategy library, and matching the target measurement error root cause with the measurement error compensation strategy library to determine the multi-scale measurement error compensation strategy. Specifically, the edge computing unit pre-constructs the measurement error compensation strategy library in its local memory. This compensation strategy library is a configuration mapping table, and each record in the table contains three fields: root cause type field, compensation scale field, and compensation algorithm parameter set field. The root cause type field stores the Chinese names of various possible measurement error root causes, such as "current transformer magnetic saturation," "voltage divider capacitor aging," "analog-to-digital converter reference voltage drift," and "temperature drift." The compensation scale field stores the compensation action dimension corresponding to the root cause, including three types: frequency scale, temperature scale, and load scale. The compensation algorithm parameter set field stores the specific algorithm name and parameter values ​​used to perform error compensation. For example, for the frequency scale, it stores the center frequency, bandwidth, and attenuation depth of the digital notch filter; for the temperature scale, it stores the polynomial order and coefficient values ​​of the temperature compensation coefficient; and for the load scale, it stores the piecewise linear fitting points and slope correction values ​​for each segment of the nonlinear compensation. After obtaining the target metrological error root cause determined in step S4333, the edge computing unit uses the Chinese name of the root cause as the query key to perform an equivalence match in the root cause type field of the metrological error compensation strategy library. If a match is successful, it reads the compensation scale and compensation algorithm parameter set from that record. The edge computing unit merges and outputs the contents of the matched compensation scale field and compensation algorithm parameter set field as a multi-scale metrological error compensation strategy. This strategy determines one or more compensation scales and their corresponding specific compensation algorithms and parameters.

[0060] Step S450: Dynamic error closed-loop compensation is performed on the sampling data of the target energy metering box based on the multi-scale metering error compensation strategy. Specifically, the edge computing unit obtains the multi-scale metering error compensation strategy determined in step S440 and loads the compensation algorithm and parameters in the strategy into the real-time compensation task module running in the edge computing unit. After each new sampling data acquisition by the target energy metering box, the real-time compensation task module immediately reads the original digital value of the sampling point and executes the compensation operation according to the compensation scale and corresponding algorithm sequence specified in the multi-scale metering error compensation strategy. If the strategy includes frequency scale compensation, the real-time compensation task module sends the sampling point data to a digital notch filter for filtering processing. The filter parameters directly adopt the center frequency, bandwidth, and attenuation depth values ​​stored in the strategy. If the strategy includes temperature scale compensation, the real-time compensation task module reads the current ambient temperature sensor value, substitutes the temperature value into the polynomial temperature compensation coefficient calculation formula stored in the strategy, calculates the compensation gain at that temperature, and multiplies the compensation gain by the filtered sampling point data. If the strategy includes load-scale compensation, the real-time compensation task module reads the current load current value, locates the current load current interval in the piecewise linear fitting points stored in the strategy, reads the slope correction value corresponding to that interval, and corrects the slope of the data after the first two compensation steps. After completing the compensation calculations for all scales, the real-time compensation task module writes the compensated digital value into the final storage area of ​​the target energy metering box, replacing the original value of that sampling point. The above compensation process is repeated for each newly acquired sampling point, forming a dynamic closed-loop compensation. The error sources determined by root cause search are precisely matched with the compensation strategy to achieve targeted multi-dimensional joint compensation, which is executed immediately after each sampling, ensuring that the metering data maintains high metering accuracy even if the error sources are not physically repaired.

[0061] This application's embodiments solve the technical problem of existing power metering box error monitoring, which cannot systematically identify anomalies in real-time operating data and trace the root cause of errors without power outages. It achieves the technical effect of systematically identifying anomalies in real-time multi-dimensional operating data and accurately locating the root cause of error propagation without power outages, thereby improving the pertinence and reliability of metering error compensation. This is achieved by acquiring multi-dimensional time-series data and performing segmented anomaly identification to filter initial anomaly segments, configuring a dual-parallel hardware verification circuit, using this circuit to cross-verify the initial anomaly segments to extract true anomaly segments, and introducing a joint search mechanism to perform root cause localization and multi-scale dynamic error compensation for true anomaly segments.

[0062] In the above text, refer to Figures 1-3 This paper describes in detail an intelligent compensation method for errors in power metering boxes based on timing anomaly detection according to embodiments of the present invention. Next, reference will be made to... Figure 4This invention describes an intelligent error compensation system for an energy metering box based on timing anomaly detection according to an embodiment of the present invention.

[0063] The intelligent error compensation system for electricity metering boxes based on time-series anomaly detection, according to embodiments of the present invention, addresses the technical problem of existing electricity metering box error monitoring systems being unable to systematically identify anomalies in real-time operating data and trace the root cause of errors without power outages. It achieves the technical effect of systematically identifying anomalies in real-time collected multi-dimensional operating data and accurately locating the root cause of error propagation without power outage intervention, thereby improving the targeting and reliability of metering error compensation. The intelligent error compensation system for electricity metering boxes based on time-series anomaly detection includes: a segmented anomaly identification module 10, a hardware verification logic circuit configuration module 20, a cross-verification module 30, and a multi-scale dynamic error compensation module 40.

[0064] The segmented anomaly identification module 10 is used to acquire multi-dimensional time-series operational data streams through the built-in acquisition terminal of the target energy metering box, and to perform segmented anomaly identification on the multi-dimensional time-series operational data streams in combination with historical operational data of the energy metering box, thereby filtering initial abnormal operational data segments; the hardware verification logic circuit configuration module 20 is used to configure hardware verification logic circuits inside the target energy metering box, wherein the hardware verification logic circuits are dual-channel parallel signal comparison channels, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of adjacent metering boxes in the same area; the cross-verification module 30 is used to perform cross-verification on the initial abnormal operational data segments using the hardware verification logic circuits to obtain the true abnormal operational data segments; the multi-scale dynamic error compensation module 40 is used to introduce a joint search mechanism to perform metering error root cause search on the true abnormal operational data segments, determine the target metering error root cause, and perform multi-scale dynamic error compensation based on the target metering error root cause.

[0065] The detailed description of the specific configuration of the segmented anomaly identification module 10 is explained as follows: As mentioned above, the segmented anomaly identification module 10, which combines historical operating data of the power metering box to identify segmented anomalies in the multi-dimensional time-series operating data stream and filters initial abnormal operating data segments, may further include: a standardized data program construction unit for constructing a standardized data program based on the power metering box data application standard, the standardized data program including time-series dimension alignment, missing data completion, and pulse anomaly point removal; a standardized processing unit for standardizing the multi-dimensional time-series operating data stream according to the standardized data program to obtain a standard multi-dimensional time-series operating data stream; an adaptive time-series segmentation unit for performing adaptive time-series segmentation of the standard multi-dimensional time-series operating data stream using a preset dynamic sliding window to obtain a set of multi-dimensional time-series operating data segments; and an anomaly identification unit for identifying anomalies in the set of multi-dimensional time-series operating data segments in combination with historical operating data of the power metering box and filtering initial abnormal operating data segments.

[0066] Specifically, the anomaly identification unit, which combines historical operating data from the electricity metering box to identify anomalies in the multidimensional time-series operating data segment set and filters initial abnormal operating data segments, may further include: a multidimensional correlation operating feature set extraction subunit for extracting multidimensional correlation operating features from the multidimensional time-series operating data segment set; a segmented data autocorrelation coefficient set calculation subunit for calculating segmented data autocorrelation coefficient sets for the same period as the multidimensional time-series operating data segment set, based on historical operating data from the electricity metering box; a dynamic feature extraction subunit for performing dynamic feature extraction on the multidimensional time-series operating data segment set based on the segmented data autocorrelation coefficient set to obtain a dynamic baseline operating feature set; and an anomaly morphology identification subunit for performing recursive feature elimination and anomaly morphology identification on the multidimensional correlation operating feature set and the dynamic baseline operating feature set to filter initial abnormal operating data segments.

[0067] The specific configuration of the cross-verification module 30 is described in detail below: As mentioned above, the hardware verification logic circuit is used to perform cross-verification on the initial abnormal operation data segment to obtain the real abnormal operation data segment. The cross-verification module 30 may further include: a signal calling unit used to call the original sampling signal of the main circuit and the synchronization reference signal of the transformer area corresponding to the initial abnormal operation data segment using the hardware verification logic circuit; a point-by-point cross-comparison unit used to perform point-by-point cross-comparison on the original sampling signal of the main circuit and the synchronization reference signal of the transformer area to obtain the deviation comparison verification result; a rejection and correction unit used to reject and correct the initial abnormal operation data segment if the deviation comparison verification result is within a preset continuous deviation range, indicating a false interference anomaly; and a marking output unit used to mark and output the initial abnormal operation data segment if the deviation comparison verification result exceeds the preset continuous deviation range, indicating a real measurement anomaly, to obtain the real abnormal operation data segment.

[0068] The detailed description of the specific configuration of the multi-scale dynamic error compensation module 40 is explained as follows: As mentioned above, a joint search mechanism is introduced to perform a root cause search for metering errors on the real abnormal operation data segment to determine the target root cause of metering errors. The multi-scale dynamic error compensation module 40 may further include: an abnormal topology analysis unit for performing abnormal topology analysis based on the metering abnormal case library of the target energy metering box to construct an abnormal link topology diagram of the energy metering box; a metering abnormality determination unit for performing metering abnormality determination by traversing the historical operation data of adjacent time periods starting from the time window of the real abnormal operation data segment to obtain directional metering abnormality type characteristics; and a metering error root cause search unit for introducing a joint search mechanism to perform a root cause search for metering errors within the abnormal link topology diagram of the energy metering box based on the directional metering abnormality type characteristics to determine the target root cause of metering errors.

[0069] The process includes: performing anomaly topology analysis based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology diagram of the energy metering box. The anomaly topology analysis unit may further include: a tagging and decomposition subunit for tagging and decomposing the metering anomaly case library of the target energy metering box to obtain a structured metering anomaly dataset; a topology node hierarchical architecture construction subunit for constructing a topology node hierarchical architecture based on the actual hardware level of the target energy metering box; an error-related edge weight assignment subunit for assigning error-related edge weights to the topology node hierarchical architecture according to the structured metering anomaly dataset to generate a basic anomaly link topology diagram; and a verification subunit for verifying and iteratively optimizing the basic anomaly link topology diagram to construct the energy metering box anomaly link topology diagram.

[0070] The method involves introducing a joint search mechanism to perform a root cause search for metering errors within the abnormal link topology of the energy metering box based on the characteristics of the directional metering anomaly type, thereby determining the target root cause of metering errors. The root cause search unit may further include: a node matching and activation subunit for performing node matching and activation on the abnormal link topology of the energy metering box according to the characteristics of the directional metering anomaly type, resulting in a queue of search nodes to be visited; a breadth-first search subunit for traversing the queue of search nodes to be visited using a joint search mechanism to perform a breadth-first search for initial screening, obtaining a set of candidate search nodes; and a depth-first search subunit for performing a depth-first search for root causes of errors within the abnormal link topology of the energy metering box based on the set of candidate search nodes, thereby determining the target root cause of metering errors.

[0071] Specifically, the error root cause depth-first search is performed within the abnormal link topology of the energy metering box based on the candidate search node set to determine the target metering error root cause. The error root cause depth-first search subunit may further include: a depth-first traversal search component used to perform a depth-first traversal search layer by layer upward within the abnormal link topology of the energy metering box, starting from the candidate search node set, to obtain a set of associated parent nodes; an anomaly determination component used to determine anomalies in the set of associated parent nodes, and if the set of associated parent nodes is an anomaly node, to continue the depth-first search layer by layer along the branches of the set of associated parent nodes to generate a feasible metering error root cause propagation chain set; and a credibility verification and screening component used to perform credibility verification and screening on the feasible metering error root cause propagation chain set to determine the target metering error root cause.

[0072] The multi-scale dynamic error compensation module 40, which performs multi-scale dynamic error compensation based on the target metering error root cause, may further include: a scenario type matching unit for constructing a metering error compensation strategy library, performing scenario type matching between the target metering error root cause and the metering error compensation strategy library, and determining a multi-scale metering error compensation strategy; and a dynamic error closed-loop compensation unit for performing dynamic error closed-loop compensation on the sampling data of the target energy metering box based on the multi-scale metering error compensation strategy.

[0073] The intelligent compensation system for power metering box errors based on timing anomaly detection provided in this embodiment of the invention can execute the intelligent compensation method for power metering box errors based on timing anomaly detection provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0074] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent compensation of errors in power metering boxes based on time-series anomaly detection, characterized in that, The method includes: The multi-dimensional time-series operation data stream is acquired through the built-in acquisition terminal of the target power metering box. Combined with the historical operation data of the power metering box, the multi-dimensional time-series operation data stream is segmented for anomaly identification, and the initial abnormal operation data segments are selected. A hardware verification logic circuit is configured inside the target energy metering box. The hardware verification logic circuit is a dual-channel parallel signal comparison channel, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of the adjacent metering box in the same area. The hardware verification logic circuit is used to perform cross-verification on the initial abnormal running data segment to obtain the real abnormal running data segment. A joint search mechanism is introduced to perform a root cause search for measurement errors on the real abnormal operation data segment, determine the target root cause of measurement errors, and perform multi-scale dynamic error compensation based on the target root cause of measurement errors.

2. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 1, characterized in that, By combining historical operating data from the electricity metering box, segmented anomaly identification is performed on the multi-dimensional time-series operating data stream to filter initial abnormal operating data segments, including: Based on the data application standards of electricity metering boxes, a standardized data program is constructed, which includes time-series alignment, missing data completion, and pulse anomaly point removal. The multidimensional time-series running data stream is standardized according to the standardized data procedure to obtain a standard multidimensional time-series running data stream. A preset dynamic sliding window is used to adaptively segment the standard multidimensional time-series running data stream to obtain a set of multidimensional time-series running data segments. By combining historical operating data from the electricity metering box, anomalies are identified in the multi-dimensional time-series operating data segment set, and initial abnormal operating data segments are selected.

3. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 2, characterized in that, By combining historical operating data from the electricity metering box, anomaly identification is performed on the multi-dimensional time-series operating data segment set, and initial abnormal operating data segments are selected, including: Extract and obtain a multidimensional associated runtime feature set of the multidimensional time-series runtime data segment set; The set of autocorrelation coefficients of segmented data within the same time period as the multidimensional time-series operational data segment set is calculated by combining historical operational data of the electricity metering box; Based on the set of autocorrelation coefficients of the segmented data, dynamic features are extracted from the set of multidimensional time-series running data segments to obtain a set of dynamic baseline running features. Recursive feature elimination and abnormal morphology identification are performed on the multidimensional associated running feature set and the dynamic baseline running feature set to filter initial abnormal running data segments.

4. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 1, characterized in that, The hardware verification logic circuit is used to perform cross-verification on the initial abnormal running data segment to obtain the real abnormal running data segment, including: The hardware verification logic circuit is used to call the original sampling signal of the main circuit and the synchronization reference signal of the transformer area corresponding to the initial abnormal operation data segment; A point-by-point cross-comparison is performed on the original sampling signal of the main circuit and the synchronous reference signal of the transformer area to obtain the deviation comparison and verification results. If the deviation comparison and verification result is within the preset continuous deviation range, it is a false interference anomaly, and the initial abnormal running data segment is removed and corrected. If the deviation comparison and verification result exceeds the preset continuous deviation range, it is a real measurement anomaly. The initial abnormal operation data segment is marked and output to obtain the real abnormal operation data segment.

5. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 1, characterized in that, A joint search mechanism is introduced to perform a root cause search for measurement errors on the actual abnormal operation data segment, and to determine the target root cause of measurement errors, including: Anomaly topology analysis is performed based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology diagram of the energy metering box; Starting from the time window of the actual abnormal operation data segment, the historical operation data of adjacent time periods are traversed to determine the metering anomaly and obtain the characteristics of the targeted metering anomaly type. A joint search mechanism is introduced to perform a root cause search for metering errors within the abnormal link topology of the power metering box based on the characteristics of the directional metering anomaly type, thereby determining the target root cause of metering errors.

6. The intelligent compensation method for errors in power metering boxes based on time-series anomaly detection as described in claim 5, characterized in that, Anomaly topology analysis is performed based on the metering anomaly case library of the target energy metering box to construct an anomaly link topology diagram of the energy metering box, including: The target energy metering box's metering anomaly case library is decomposed into a tagged dataset to obtain a structured metering anomaly dataset; Based on the actual hardware hierarchy of the target power metering box, a topology node hierarchical architecture is constructed. Based on the structured measurement anomaly dataset, error-related edge weights are assigned to the hierarchical architecture of the topology nodes to generate a basic anomaly link topology graph; The basic abnormal link topology is verified and iteratively optimized to construct an abnormal link topology for the power metering box.

7. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 5, characterized in that, A joint search mechanism is introduced to perform a root cause search for metering errors within the abnormal link topology of the energy metering box based on the characteristics of the directional metering anomaly type, to determine the target root cause of the metering error, including: Based on the characteristics of the directional metering anomaly type, the abnormal link topology of the power metering box is matched and activated to obtain a queue of nodes to be accessed and searched. A joint search mechanism is introduced to traverse the queue of search nodes to be visited and perform a breadth-first search for initial screening to obtain a set of candidate search nodes. Based on the set of candidate search nodes, perform a depth-first search for the root cause of the error within the abnormal link topology of the power metering box to determine the target metering error root cause.

8. The intelligent compensation method for power metering box errors based on time-series anomaly detection as described in claim 7, characterized in that, Based on the candidate search node set, a depth-first search for the root cause of the error is performed within the abnormal link topology of the power metering box to determine the target metering error root cause, including: Starting from the set of candidate search nodes, a depth-first traversal search is performed layer by layer upwards within the abnormal link topology of the power metering box to obtain the set of associated parent nodes; Anomaly determination is performed on the set of associated parent nodes. If the set of associated parent nodes is an anomaly, the depth search continues layer by layer along the branches of the set of associated parent nodes to generate a feasible set of measurement error root cause transmission chains. The feasible measurement error root cause propagation chain set is subjected to credibility verification and screening to determine the target measurement error root cause.

9. The intelligent compensation method for errors in power metering boxes based on time-series anomaly detection as described in claim 1, characterized in that, Multi-scale dynamic error compensation based on the target measurement error root cause includes: A measurement error compensation strategy library is constructed, and a multi-scale measurement error compensation strategy is determined by matching the target measurement error root cause with the measurement error compensation strategy library for different scenarios. Based on the multi-scale metering error compensation strategy, dynamic error closed-loop compensation is performed on the sampling data of the target power metering box.

10. An intelligent compensation system for errors in an energy metering box based on time-series anomaly detection, characterized in that, The system is used to implement the intelligent compensation method for power metering box errors based on time-series anomaly detection as described in any one of claims 1-9, and the system comprises: The segmented anomaly identification module is used to acquire multi-dimensional time-series operation data streams through the built-in acquisition terminal of the target power metering box, and to perform segmented anomaly identification on the multi-dimensional time-series operation data streams in combination with the historical operation data of the power metering box, and to filter the initial abnormal operation data segments. The hardware verification logic circuit configuration module is used to configure a hardware verification logic circuit inside the target energy metering box. The hardware verification logic circuit is a dual-channel parallel signal comparison channel, wherein the first channel is connected to the sampling signal of the metering main circuit, and the second channel is connected to the synchronization reference signal of the adjacent metering box in the same area. The cross-validation module is used to perform cross-validation on the initial abnormal running data segment using the hardware verification logic circuit to obtain the real abnormal running data segment. The multi-scale dynamic error compensation module is used to introduce a joint search mechanism to perform a root cause search for measurement errors on the real abnormal operation data segment, determine the target root cause of measurement errors, and perform multi-scale dynamic error compensation based on the target root cause of measurement errors.

Images