Power metering drift correction method and system based on transformer area line loss and medium

Abstract

The application discloses a power metering drift correction method and system based on transformer area line loss and a medium, relates to the technical field of power meters, and comprises the following steps: constructing a four-level topology of a transformer area-branch-meter box-meter and attaching a communication quality label; calculating a measured line loss rate in layers, combining a physical constraint graph neural network to obtain a theoretical line loss; comparing a deviation and combining communication quality to determine suspicious nodes; performing time sequence decomposition on candidate meters to identify drift types; and adaptively verifying and grading correcting according to the drift types. The application solves the technical problems that it is difficult to determine the power metering drift node in the prior art, the drift type cannot be effectively distinguished, the operation and maintenance efficiency is low, and drift identification is prone to errors, and achieves the technical effects of improving line loss analysis accuracy by constructing a refined topology and a graph neural network, accurately identifying the drift type by combining a communication quality label and time sequence decomposition, and improving the reliability of power metering drift processing and the operation and maintenance efficiency.

Description

Technical Field

[0001] This invention relates to the field of electricity meter technology, and in particular to a method, system and medium for correcting meter drift based on transformer area line loss. Background Technology

[0002] As the core equipment for metering and settlement in the power system, the accuracy of electricity meters directly affects the economic interests of both power suppliers and consumers. With long-term operation, electricity meters are affected by factors such as component aging, temperature changes, and load fluctuations, which can cause metering drift in the metering chip or sampling circuit, leading to a gradual increase in metering errors. This can result in problems such as abnormal line losses in distribution areas and disputes over electricity bill collection. Traditional methods for detecting metering errors mainly rely on periodic on-site calibration or manual sampling. On-site calibration requires power outages or bringing standard meters for testing, which has drawbacks such as high labor costs, long testing cycles, and limited coverage, making it difficult to detect early metering drift in a timely manner. With the increasing demand for comprehensive coverage of electricity information collection systems and refined management of distribution area line losses, using distribution area line loss data for metering anomaly analysis has gradually become a research hotspot.

[0003] Existing line loss-based electricity meter anomaly detection typically uses a fixed threshold for line loss rate judgment, marking nodes with line loss rates exceeding a preset range as abnormal. However, actual transformer area line losses are affected by various factors such as line impedance, load fluctuations, ambient temperature, and topology, making it difficult to accurately distinguish between metering drift and normal line loss fluctuations using a single fixed threshold. Furthermore, most existing methods directly mark abnormal points as electricity meter faults, failing to deeply identify the specific type of metering drift, resulting in a simplistic subsequent processing strategy and hindering accurate correction and efficient operation and maintenance. Summary of the Invention

[0004] This invention provides a method, system, and medium for correcting meter drift based on transformer substation line loss. The key aspect lies in addressing the technical obstacles in transformer substation line loss analysis scenarios, such as the difficulty in accurately identifying and locating meter drift, high misjudgment rates, and a lack of targeted correction strategies due to the characteristics of multi-source heterogeneous data acquisition with communication quality differences, complex topological relationships, and strong time-varying metering errors. This is achieved by using a graph neural network that integrates a physical model of line loss for theoretical line loss modeling and hierarchical deviation analysis. Combined with time series decomposition, an algorithm for drift type identification is adapted. Furthermore, a data processing flow based on a four-level topology (transformer substation-branch-meter box-meter) and communication quality tags is employed to achieve high-precision identification and classification of meter drift, rapid location of abnormal nodes, and hierarchical adaptive correction. This improves the accuracy of line loss analysis and reduces operation and maintenance costs.

[0005] In a first aspect, the present invention provides a method for correcting meter drift based on transformer substation line loss, wherein the method for correcting meter drift based on transformer substation line loss includes:

[0006] A four-level hierarchical topology structure of transformer area-branch-meter box-meter is constructed, and a communication quality label is attached to each data collection. Based on the hierarchical topology, the measured line loss rate is calculated layer by layer, and the theoretical line loss benchmark value is calculated using a graph neural network with an embedded physical model of line loss. The deviation between the measured line loss rate and the theoretical line loss benchmark value is calculated, and the communication quality label of the corresponding data is read to identify suspicious drift candidate nodes. For the suspicious drift candidate meters corresponding to the suspicious drift candidate nodes, the historical error change trajectory time series decomposition is performed to identify the drift type. Based on the identified drift type, adaptive verification processing is performed to determine the verified drift type and error magnitude, and a hierarchical correction strategy is executed. The hierarchical correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

[0007] Secondly, the present invention also provides a metering drift correction system for electricity meters based on transformer area line loss, wherein the metering drift correction system for electricity meters based on transformer area line loss includes: Topology Construction Module: Constructs a four-level hierarchical topology structure of transformer area-branch-meter box-meter, and attaches a communication quality label to each collected data. Theoretical Line Loss Calculation Module: Calculates the measured line loss rate layer by layer based on the hierarchical topology structure, and calculates the theoretical line loss benchmark value using a graph neural network embedded with a physical model of line loss. Line Loss Deviation Identification Module: Calculates the deviation between the measured line loss rate and the theoretical line loss benchmark value, and reads the communication quality label of the corresponding data to identify suspicious drift candidate nodes. Time Series Decomposition Module: Performs time series decomposition of the historical error change trajectory of the suspicious drift candidate meters corresponding to the suspicious drift candidate nodes to identify the drift type. Hierarchical Correction Module: Based on the identified drift type, performs adaptive verification processing to determine the verified drift type and error magnitude, and executes a hierarchical correction strategy. The hierarchical correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

[0008] Thirdly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the metering drift correction method for electricity meters based on transformer area line loss provided by the present invention.

[0009] This invention discloses a method, system, and medium for correcting meter drift based on transformer substation line loss, comprising: constructing a four-level hierarchical topology structure of transformer substation-branch-meter box-meter, and attaching a communication quality tag to each piece of collected data; calculating the measured line loss rate layer by layer based on the hierarchical topology structure, and calculating the theoretical line loss benchmark value using a graph neural network embedded with a physical model of line loss; calculating the deviation between the measured line loss rate and the theoretical line loss benchmark value, and reading the communication quality tag of the corresponding data to determine suspected drift candidate nodes; performing time series decomposition of the historical error change trajectory of the suspected drift candidate meters corresponding to the suspected drift candidate nodes to identify the drift type; and performing adaptive verification processing based on the identified drift type. The present invention discloses a method, system, and medium for correcting electricity meter drift based on transformer substation line loss. This method determines the drift type and error magnitude, and executes a tiered correction strategy. The tiered correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks. It solves the technical problems in existing technologies, such as difficulty in accurately locating electricity meter drift nodes and ineffective differentiation of drift types, leading to low maintenance efficiency and high misjudgment of drift identification. The method achieves the technical effect of improving the accuracy of line loss analysis by constructing a refined topology and graph neural network, and accurately identifying drift types by combining communication quality tags and time series decomposition, thereby improving the reliability of electricity meter drift processing and the efficiency of maintenance. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the metering drift correction method for electricity meters based on transformer substation line loss according to the present invention.

[0011] Figure 2 This is a schematic diagram of the energy meter drift correction system based on transformer substation line loss according to the present invention.

[0012] Figure labeling: Topology construction module 11, theoretical line loss calculation module 12, line loss deviation identification module 13, time series decomposition module 14, graded correction module 15. Detailed Implementation

[0013] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0014] Example 1, as Figure 1This is a flowchart illustrating the method for correcting meter drift based on transformer substation line loss according to the present invention. The method includes: A four-level hierarchical topology of transformer area-branch-meter box-meter is constructed, and a communication quality label is attached to each piece of collected data.

[0015] Specifically, the electrical connections within the target transformer area are first meticulously analyzed using current pulses. Using the main transformer area meter as the root node, power supply branches are sequentially divided downwards, further refined to the meter boxes corresponding to each branch, and finally located at the specific user's electricity meter. This establishes a four-level hierarchical relationship: transformer area - branch - meter box - electricity meter. Simultaneously, a unique identifier is assigned to each node, and the connection relationships between parent and child nodes, electrical affiliation, and topology hierarchy information are recorded. This constructs a hierarchical topology structure suitable for calculation and analysis. This hierarchical topology not only reflects physical connections but also supports subsequent layered line loss calculations and anomaly localization. After topology construction, the collected data from each electricity meter undergoes standardized access and extended processing. Based on the original collected data, a communication quality tag is added to each data record, including acquisition time, communication latency, and recall flags, to characterize the reliability of the data acquisition process. Through this mechanism, the reliability of data acquisition can be quantified at the data level, providing a reliable basis for subsequent line loss calculations, anomaly screening, and drift identification, thereby avoiding misjudgments caused by communication problems.

[0016] In some embodiments, a four-level hierarchical topology of transformer area-branch-meter box-meter is constructed, and a communication quality label is attached to each piece of collected data, including: A characteristic current pulse with preset amplitude, frequency, and time encoding is injected into the transformer area. Based on the response characteristics of each sub-meter to the characteristic current pulse, the affiliation between the meter and the branch / meter box is automatically identified, generating a four-level hierarchical topology structure of transformer area-branch-meter box-meter. The power consumption data, voltage and current curves, and operating status data of each sub-meter are collected, and a communication quality tag is attached to each collected data. The communication quality tag includes at least the collection time, communication delay, and recall flag. The recall flag is generated as follows: when the meter data cannot be successfully read within the collection period, and the data is supplemented from the cache or historical data, the recall flag of this data is set to true, and the number of recalls is recorded. When the number of consecutive recalls exceeds a preset recall threshold, the meter is marked as a communication abnormal meter.

[0017] Specifically, the concentrator or distribution terminal on the transformer substation side first injects a characteristic current pulse signal with specific parameters into the low-voltage lines of the target transformer substation during a preset low-load period, such as the off-peak hours at night. This characteristic current pulse includes a preset amplitude, a preset frequency, and a unique time code. The preset amplitude is used to ensure that the characteristic current pulse can be reliably sensed and identified by each energy meter without affecting normal power consumption and metering accuracy. The preset frequency is used to distinguish it from power frequency and load fluctuation signals to improve the identifiability and anti-interference capability of the characteristic signal. The unique time code is used to distinguish different injection batches. Subsequently, this characteristic pulse signal propagates along the distribution line to each branch, meter box, and end energy meter. Each energy meter or its associated acquisition terminal then samples and records this characteristic current pulse, extracting its response characteristics, including but not limited to pulse arrival time, response amplitude, waveform similarity, duration, and phase characteristics. After the main station aggregates the response data of each meter, it automatically infers the electrical connection relationship between the meters based on the timing relationship of pulse propagation, response consistency and signal attenuation law, thereby determining the meter box to which each meter belongs, the branch to which each meter box belongs, and the transformer area to which each branch belongs. Finally, a four-level hierarchical topology structure of transformer area-branch-meter box-meter is constructed and stored in the form of a graph structure.

[0018] After completing the topology construction, the system enters the routine data acquisition phase, periodically reading multi-dimensional operational data from each electricity meter through the electricity information acquisition unit, including electricity consumption data, voltage and current curve data, and operational status data. For each acquired data record, the system simultaneously generates and attaches a communication quality tag upon entry into the database. When generating the communication quality tag, the system first records the acquisition time and the master station reception time of the data, and calculates the communication latency by the difference between the master station reception time and the acquisition time. At the same time, it marks the data acquisition method, i.e., determines whether the data is the first successfully read data within the acquisition cycle. If the data is successfully acquired within the specified acquisition cycle, such as within 15 minutes or 1 hour, the recall flag is set to false. If the data is not successfully acquired within the cycle but is subsequently acquired through terminal caching, concentrator caching, or historical data supplementation, the recall flag for that data is set to true, the recall behavior is recorded in the tag, and the recall counter of the corresponding electricity meter is incremented by 1. Subsequently, the system counts the number of consecutive recalls for each meter. If a meter relies on recalls to acquire data in multiple consecutive data collection cycles, and its cumulative recall count exceeds a preset threshold (e.g., 3 or 5 consecutive times), the system determines that the meter's communication link is abnormal, marks it as a communication anomaly meter, and can trigger subsequent communication maintenance or data removal strategies. Through these steps, a complete process is achieved, from injecting feature signals to identify topology relationships, to multi-source data acquisition, and then to fine-grained labeling of communication quality tags, providing a highly reliable data foundation for subsequent line loss calculation and metering drift identification.

[0019] For example, when performing a topology identification on the low-voltage side of a 10kV distribution substation, the low-load period at 02:30 am is selected, and a set of characteristic current pulse signals is injected by the substation concentrator: the pulse reference current amplitude is set to 2% of the rated current of 100A, then the injected pulse amplitude is 2A, the pulse frequency is set to 75Hz, which is different from the 50Hz power frequency, the pulse duration is 200ms, the pulse interval is 800ms, and a unique time code 20240402-0230-01 is assigned to this injection, and a coding sequence of 3 short and 1 long is used, that is, 3 consecutive 200ms short pulses followed by 1 400ms long pulse as an identifier. Each electricity meter locally records the current waveform at a sampling rate of 1kHz. Corresponding pulse signals are detected at 02:30:15.200s, 02:30:16.000s, 02:30:16.800s, and 02:30:17.800s, with recorded response amplitudes ranging from approximately 1.6A to 2.1A and phase shifts from approximately 3° to 8°. After aggregation by the main station, the meter boxes and branch nodes closest to the power supply are determined based on the timing and amplitude attenuation of the pulses detected by each meter, thus automatically identifying the meter hierarchy within this batch of distribution areas.

[0020] The measured line loss rate is calculated in layers based on the hierarchical topology, and the theoretical line loss benchmark value is calculated using a graph neural network embedded with the physical model of line loss.

[0021] Specifically, after completing the hierarchical topology modeling of the transformer substation-branch-meter box-meter, the system first performs layered line loss calculations from top to bottom based on this topology. In this process, the input power of any parent node is used as a benchmark, and the sum of the power of all its corresponding child nodes is used as the output. The difference between the two is normalized to obtain the measured line loss rate for that level. By repeating the above calculation process at different levels, a multi-level line loss distribution from the transformer substation to the end meter can be formed, thereby achieving a progressively refined analysis and location of line losses. Based on the obtained measured line losses, to improve the accuracy of anomaly detection, a graph neural network model incorporating line physical characteristics is introduced to calculate the theoretical line loss benchmark value. Specifically, the system transforms the constructed hierarchical topology into a graph structure, where various types of nodes are nodes in the graph, and the electrical connections between nodes are edges. Multi-dimensional features are assigned to nodes and edges respectively; for example, the node side includes historical power, voltage, and current, while the edge side includes information such as line length, conductor type, and equivalent impedance. Based on this, the aforementioned graph structure and its characteristics are input into a graph neural network model. Combined with the physical constraints of line loss included in the physical model of line loss, a comprehensive modeling of the electrical state of each node is achieved through multi-layer information propagation and aggregation. The physical model of line loss is embedded within the graph neural network model, ensuring that the model output not only conforms to data distribution characteristics but also satisfies the physical laws of the power system. Finally, the graph neural network model outputs a corresponding theoretical line loss benchmark value for each node, serving as a reference standard under ideal conditions. By comparing the measured line loss rates at each level with the theoretical line loss benchmark values ​​calculated by the graph neural network, abnormal deviations can be identified more accurately, providing a reliable basis for subsequent drift detection and location, thereby improving the accuracy of line loss analysis and reducing the false positive rate.

[0022] In some embodiments, the measured line loss rate is calculated hierarchically based on the hierarchical topology, and the theoretical line loss benchmark value is calculated using a graph neural network embedded in the physical model of line loss, including: Based on the hierarchical topology, the ratio of the difference between the input power of the parent node and the output power of the child node to the input power of the parent node is calculated to obtain the measured line loss rate of each node. The hierarchical topology is mapped to a graph structure, where nodes correspond to main meters, branch nodes, meter box nodes, and energy meter nodes, and edges correspond to parent-child connection relationships. Using node features and edge features as input, a graph neural network model is used for forward propagation to output the theoretical line loss baseline value of each node. The node features include at least node type, historical power, real-time voltage, real-time current, power factor, and load fluctuation characteristics. The edge features include at least line length, conductor type, equivalent impedance, real-time current, ambient temperature, and topology level. The graph neural network model has physical constraints on line loss, which include at least line ohmic loss constraints and node power conservation constraints.

[0023] Specifically, within a preset statistical period, according to the established topology of transformer area-branch-meter box-electricity meter hierarchy, the input electricity of each parent node is read as the input electricity. The output electricity of all direct child nodes of the parent node within the same statistical period is then summarized. The difference between the input electricity of the parent node and the sum of the output electricity of each child node is calculated, and this difference is divided by the input electricity of the parent node to obtain the measured line loss rate of the parent node in the current statistical period. Specifically, for the transformer area level, the input electricity of the transformer area main meter or the total metering device at the low-voltage outlet of the distribution transformer is used as the input electricity of the parent node, and the sum of the total electricity of each branch node is used as the sum of the output electricity of the child nodes. For the branch level, the input electricity of the branch entrance is used as the input electricity of the parent node, and the sum of the total electricity of each meter box node under the branch is used as the sum of the output electricity of the child nodes. For the meter box level, the total input electricity of the meter box is used as the input electricity of the parent node, and the sum of the metered electricity of each electricity meter connected to it is used as the sum of the output electricity of the child nodes. Before performing the above calculations, the data collected from each node is preprocessed in a unified manner, including timestamp alignment, missing value completion, outlier removal, and unit conversion. Data with too many recall attempts, excessive communication latency, or marked as communication anomalies can be removed, downweighted, or replaced with estimated values ​​from adjacent periods to ensure that the measured line loss rate calculation results are comparable and reliable.

[0024] Subsequently, the hierarchical topology is mapped to graph structure data that can be processed by a graph neural network. Here, the main meter node, branch nodes, meter box nodes, and energy meter nodes correspond to different types of nodes in the graph. The parent-child power supply relationship between nodes corresponds to edges in the graph, with the edge direction preferably set from the upper-level power supply node to the lower-level power receiving node. Simultaneously, a corresponding feature vector is constructed for each node and each edge. The node feature vector includes at least node type, historical energy consumption, real-time voltage, real-time current, power factor, and load fluctuation characteristics. The node type characterizes whether the node belongs to the main meter, branch meter, meter box, or energy meter category. Historical energy consumption reflects the average load level and variation of the node within the historical statistical window. The dynamic law, real-time voltage, real-time current and power factor are used to characterize the operating status under the current working conditions. The load fluctuation characteristics can be obtained from the standard deviation, range, rate of change or coefficient of variation within a preset time window. The edge feature vector includes at least the line length, conductor type, equivalent impedance, real-time current, ambient temperature and topology level. The line length and conductor type are obtained from the operation and maintenance archives or on-site surveys. The equivalent impedance is calculated based on the line length, conductor material parameters and cross-sectional area. The real-time current is taken as the current transmission current of the power supply line corresponding to the edge. The ambient temperature is taken as the temperature sensor data or meteorological data of the area where the line is located. The topology level is used to characterize the connection relationship of the edge from the transformer area to the branch, from the branch to the meter box or from the meter box to the meter.

[0025] After constructing node and edge features, the graph structure is input into a pre-trained graph neural network model for forward propagation. During message passing at each layer, information from the upper-level node and adjacent edges is aggregated to the lower-level node or the current node to learn the power transmission relationship and loss distribution law between nodes at different levels. Finally, the theoretical line loss benchmark value corresponding to each node is output. To avoid the model relying solely on statistical fitting and ignoring the objective physical laws of the power grid, the graph neural network model embeds a line loss physical model that includes physical constraints on line losses. These physical constraints include line ohmic loss constraints and node power conservation constraints. The line ohmic loss constraints are used to ensure that the theoretical loss on each edge satisfies the relationship related to the line resistance and the square of the current, i.e., the edge loss follows the I²R loss law. The node power conservation constraints are used to ensure that the input power or input quantity of any parent node should be equal to the sum of the output power or output quantity of each child node and the approximate balance relationship between the sum of the line losses from the parent node to each child node. In practical implementation, the aforementioned ohmic loss constraint and node power conservation constraint can be jointly incorporated into the model loss function. During model training, both the theoretical line loss prediction error and the physical constraint deviation can be minimized simultaneously. This ensures that the output theoretical line loss benchmark value conforms to both historical sample patterns and the actual operating mechanism of distribution lines. Finally, the system associates and stores the theoretical line loss benchmark value of each node with the measured line loss rate under the corresponding statistical period, providing basic data for subsequent deviation analysis, identification of suspected drifting nodes, and metering correction.

[0026] When training the graph neural network model, sample data from multiple transformer substations under different time windows are first extracted from historical operating data. A corresponding graph structure is constructed for each sample, including a set of nodes, edge connections, and corresponding node and edge features. Simultaneously, supervision labels are generated based on historical stable operating intervals or manually verified data, typically representing the reference line loss level of each node in the corresponding time period. Next, the node and edge features are normalized, missing values ​​are imputed, and model parameters are initialized, including node embedding dimension, message passing layers, weight matrix, bias term, and activation function. The node embedding dimension can be set to a fixed-length vector based on node complexity. The number of message passing layers is preferably 2-4 layers to balance expressive power and computational complexity. The activation function is either ReLU or LeakyReLU. Subsequently, in each training round, the constructed graph structure is input into the model, and forward propagation is performed through a message passing mechanism. Information about adjacent nodes and edges is aggregated layer by layer to obtain the implicit representation of each node, and the corresponding theoretical line loss prediction value is output. Then, the loss function is calculated. This loss function consists of two parts: one is the error loss between the prediction result and the supervision label, used to measure the model's fitting ability; the other is the physical constraint loss, used to measure the degree of deviation of the model output from the ohmic loss law of the line and the node power conservation relationship. The two parts of the loss can be weighted and fused according to preset weights. Based on this, the gradient of the parameters of each layer is calculated using the backpropagation algorithm, and an optimizer (Adam or RMSProp) is used to iteratively update the model parameters. Simultaneously, the learning rate (1e-3 to 1e-4), batch size, and regularization parameters are set to control training stability and prevent overfitting. During training, a training set and a validation set can be divided, and the model's performance on the validation set is periodically evaluated. Training stops when the validation error no longer decreases or reaches the preset convergence condition, and the optimal parameter combination is selected as the final model. Finally, the trained model is deployed to the system to perform inference calculations on the real-time or near-real-time constructed transformer area diagram structure data, output the theoretical line loss baseline value of each node, and can perform periodic incremental training or parameter fine-tuning based on new data to continuously improve the model's generalization ability and prediction accuracy under different transformer areas and different operating conditions.

[0027] Calculate the deviation between the measured line loss rate and the theoretical line loss benchmark value, and read the communication quality label of the corresponding data to identify suspected drift candidate nodes.

[0028] Specifically, after obtaining the measured line loss rate and corresponding theoretical line loss benchmark value for each level of node, the relative deviation between the measured line loss rate and the theoretical line loss benchmark value is calculated for each node to characterize the degree of deviation of the node's current operating state from the ideal state. When this deviation exceeds a pre-set threshold range, it indicates that the node may have an abnormal situation, such as measurement error or abnormal loss. Subsequently, a communication quality label corresponding to the node's data is introduced to determine whether the data is real-time reliable acquisition data: if the data is acquired in real time and the communication latency is within the normal range, and there is no recall, the data is considered to have high reliability. If the deviation is abnormal, the node can be marked as a suspected drift candidate node and enter the subsequent analysis process. Conversely, if the communication quality label shows that the data is recall data or there is excessive communication latency, it indicates that the data may have distortion or lag problems. For such nodes, even if their line loss deviation is abnormal, they are first classified as communication suspicious nodes, rather than immediately judged as measurement drift problems, and can be re-evaluated based on new real-time data in subsequent acquisition cycles. By combining line loss deviation assessment with communication quality information, it is possible to effectively distinguish between real anomalies caused by metering drift and false anomalies caused by communication problems, thereby improving the accuracy of suspicious node screening and reducing the false positive rate.

[0029] In some embodiments, the deviation between the measured line loss rate and the theoretical line loss benchmark value is calculated, and the communication quality label of the corresponding data is read to determine the suspected drift candidate nodes, including: If the deviation between the measured line loss rate and the theoretical line loss benchmark value exceeds a preset range, and the communication quality label shows that the data is collected in real time and the communication latency does not exceed the standard, then the corresponding node will be marked as a suspected drift candidate node; if the communication quality label shows that the data is supplementary data or the communication latency exceeds the standard, then the corresponding node will be marked as a communication suspicious node, and will not enter the drift type identification process for the time being, and will wait for at least one collection cycle to re-evaluate the drift suspicion based on the newly collected data.

[0030] Specifically, after calculating the measured line loss rate and theoretical line loss benchmark value for each node, the deviation value between the two is calculated for each node within the current statistical period. This deviation value is then compared with a pre-set deviation threshold range, which can be set separately for different node levels. When the absolute value of the deviation for a node exceeds the deviation threshold range, the system reads the communication quality tag corresponding to the data involved in the line loss calculation for that node to verify the data's reliability. It determines whether the data acquisition method for that node is real-time acquisition (i.e., the recall flag is false) and whether the data was successfully acquired for the first time within the specified acquisition period. Simultaneously, it calculates whether the communication latency is less than a preset communication latency threshold and determines whether the data timestamps are continuous and the fields are complete. If all the above communication quality conditions are met—that is, the data is acquired in real-time and the communication latency does not exceed the standard—the deviation of that node is considered to have high reliability. At this point, the node is marked as a suspected drift candidate node, and its relevant information is recorded in the candidate list, proceeding to the subsequent drift type identification process. Conversely, if the data communication quality label corresponding to the node indicates a recall situation (i.e., the recall flag is true), or the communication latency exceeds a preset threshold, or there are situations such as exceeding the limit for consecutive recalls or discontinuous timestamps, then it is considered that the current deviation of the node may be caused by abnormal data acquisition rather than actual metering drift. In this case, the system marks the node as a communication suspicious node and sets a temporary processing flag for it, preventing it from entering the drift type identification process. For objects marked as communication suspicious nodes, the system includes them in the delay evaluation queue and continuously monitors their data quality status for at least one subsequent acquisition cycle. After the arrival of a new acquisition cycle, the real-time acquisition data of the node is reacquired, and the line loss deviation calculation and communication quality verification process is repeated. If, in a subsequent cycle, the node's data returns to real-time acquisition and the communication latency is normal, but the deviation still continues to exceed the preset range, it is re-evaluated and upgraded to a suspicious drift candidate node to avoid misjudgment due to single data fluctuations or occasional communication anomalies, thereby improving the accuracy and robustness of candidate node selection.

[0031] For the suspected drift candidate meters corresponding to the suspected drift candidate nodes, perform time series decomposition of historical error change trajectories to identify the drift type.

[0032] Specifically, after identifying potential drift candidate nodes, the corresponding electricity meters under these nodes are further located, and their error data during historical operation is extracted to construct a time series of error change trajectories over time. Subsequently, the error change trajectory time series is decomposed into different components, including long-term trend terms, periodic seasonal terms, and randomly fluctuating residual terms. Then, based on the characteristics of each component obtained from the decomposition, the drift type of the electricity meter is determined. For example, when the trend term shows slow and monotonous changes, it can be considered a gradually accumulating drift, denoted as component aging drift; when the error change is highly correlated with temperature, it can be judged as a drift influenced by external factors, denoted as temperature-coupled drift. Through the above time series decomposition and feature analysis of historical error trajectories, a refined distinction between different types of metering drift can be achieved, providing a basis for subsequent targeted verification and differentiated correction, thereby improving the overall accuracy and processing efficiency of identification.

[0033] In some embodiments, the historical error change trajectory time series decomposition is performed on the suspected drift candidate meters corresponding to the suspected drift candidate nodes to identify the drift type, including: The historical error change trajectory time series of the identified suspected drift candidate meters is decomposed to determine the trend term, seasonal term, and random term. If the trend term is monotonic and the absolute value of the rate of change is less than a first rate of change threshold, it is determined to be component aging type drift. If the correlation coefficient between the seasonal term and the ambient temperature data is greater than a first correlation coefficient threshold, it is determined to be temperature-coupled type drift. If the random term has a step change and the step amplitude is greater than a preset amplitude threshold, it is determined to be sudden failure type drift. If the error of the suspected drift candidate meter increases significantly under low current conditions and has a nonlinear relationship with the current magnitude, it is determined to be light-load nonlinear type drift.

[0034] Specifically, for identified suspected drifting candidate meters, the system extracts their error change trajectory data within a preset historical time window and samples them at uniform time intervals to form a continuous time series. This time series is then preprocessed, including time alignment, missing value imputation, outlier removal, and low-confidence data filtering based on communication quality tags, to obtain a stable and analyzable historical error change trajectory time series. Subsequently, a time series decomposition algorithm is used to decompose the error series into three parts: a trend term, a seasonal term, and a random term. The trend term reflects the long-term trend of error changes; the seasonal term reflects periodic fluctuations; and the random term reflects sudden disturbances or irregular fluctuations. This decomposition can be implemented using methods such as STL decomposition, moving average decomposition, or wavelet decomposition. Taking STL decomposition as an example, the key parameters of STL decomposition are first set, including the seasonal period length (e.g., with a 24-hour period, the period length is 24 or 96 sampling points), the trend smoothing window length, and the seasonal smoothing window length. The trend term T(t), seasonal term S(t), and residual term R(t) are then initialized. Next, the initial trend term is extracted from the historical error trajectory time series using the Locally Weighted Regression (LOESS) method. This involves low-pass smoothing the series to obtain the trend component reflecting long-term changes. The trend term is then subtracted from the historical error trajectory time series to obtain the detrended series. This detrended series is then grouped according to periodic positions, such as historical points at the same time. Each group is smoothed again using LOESS to extract the seasonal term S(t). Afterward, the trend and seasonal terms are subtracted from the historical error trajectory time series to obtain the residual term R(t), i.e., the random term. Based on this, a robust weighting mechanism can be introduced to reduce the weight of outliers with large residuals. The trend and seasonal extraction processes are repeated several times until each component converges or its change falls below a preset threshold. Finally, a three-part decomposition result satisfying the relationship E(t) = T(t) + S(t) + R(t) is obtained, where E(t) is the historical error trajectory time series.

[0035] After obtaining each component, the system extracts and distinguishes the corresponding features. During this process, monotonicity detection and rate of change calculation are performed on the trend term. The rate of change can be obtained by performing first-order difference or linear fitting on the trend term sequence. When the trend term as a whole exhibits monotonic change and its absolute rate of change is less than a preset first rate of change threshold, it indicates that the error drifts slowly and steadily over time, consistent with component aging characteristics, thus identifying it as component aging-type drift. Correlation analysis is performed on the seasonal term and ambient temperature data, using Pearson or Spearman correlation coefficients to calculate the degree of correlation. When the correlation coefficient is greater than a preset first correlation coefficient threshold, it indicates that the error change is significantly affected by temperature, thus identifying it as temperature-coupled drift. For the random term... Sudden change detection can be performed using methods such as sliding window mean difference, CUSUM algorithm, or step detection to identify significant step changes. When a sudden change is detected in a random term and its amplitude exceeds a preset threshold, it indicates a sudden jump in error, typically corresponding to equipment failure or abnormal conditions, thus classifying it as a sudden fault-type drift. Furthermore, the relationship between error and current is analyzed, extracting the correlation between error and current values ​​across different load ranges, with a focus on low-current conditions, such as ranges below a certain percentage of the rated current. If the error is significantly amplified in low-current ranges and exhibits a clear nonlinear relationship with current changes, such as a quadratic curve or piecewise variation, it is classified as a light-load nonlinear drift. Through this multi-dimensional decomposition and discrimination process, refined identification of different types of metering drift can be achieved, providing a basis for subsequent differentiated verification and targeted correction, thereby improving the overall accuracy and reliability of the processing.

[0036] In some embodiments, identifying the drift type further includes: Establish a cross-regional meter batch database to record the error statistics of meters in the same batch across multiple regions. When the error trend terms of multiple meters in the same region are consistent in direction and have similar rates of change, and the bus loss rate of that region does not exceed the abnormal threshold, the cross-regional meter batch database is called for horizontal comparison. If meters in the same batch show the same error trend in other regions, it is determined to be a synchronous drift type drift.

[0037] Specifically, a cross-regional meter batch database is first established. This database uses meter production batch number, model specifications, manufacturer, manufacturing date, firmware version, etc., as index fields to collect and manage meters of the same batch deployed in different regions. It continuously records the error statistics of each meter during operation, including at least the historical error mean, error standard deviation, error trend direction, trend change rate, seasonal amplitude, and random fluctuation intensity. After the system completes the time series decomposition of candidate meters within a target region, if it finds that multiple meters within the same region have the same error trend direction (e.g., all showing a continuous increase or decrease), and the difference between the corresponding trend change rates of each meter is less than a preset similarity threshold, it indicates that these meters may have a collective, consistent offset. In this case, the system further reads the bus loss rate of the region within the corresponding analysis period to determine whether it exceeds a preset abnormal threshold. If the bus loss rate of the transformer area does not exceed the abnormal threshold, it indicates that the overall energy balance of the transformer area is basically normal, and does not meet the typical characteristics of local electricity theft, abnormal wiring, or single-point fault causing large-scale loss anomalies. Therefore, the system regards the current consistency trend of multiple meters as a candidate for batch common problems, rather than transformer area topology loss anomalies. Subsequently, based on the batch number, model specifications, etc. of the currently suspected meters, the system retrieves samples of meters from the same batch deployed in other transformer areas from the cross-transformer area meter batch database, and then selects control samples with similar operating conditions to the current transformer area. The selection criteria include similar load type, similar ambient temperature range, similar operating time, and similar communication quality. Next, the error trend terms of other transformer substations in the same batch are compared and analyzed to determine whether the trend direction is consistent, whether the trend change rate falls within the preset similarity range, whether the trend duration reaches the minimum duration threshold, and whether the trend consistency ratio among multiple samples exceeds the preset consistency ratio threshold. For example, it can be set that when no less than a preset proportion of transformer substations in the same batch show the same upward or downward trend as the current target transformer substation, and the difference in the rate of change does not exceed the allowable deviation range, it is considered that there is a cross-transformer substation batch consistency drift phenomenon.

[0038] Furthermore, to avoid misjudging general operating condition changes as synchronous drift, the system can add exclusionary checks, i.e., checking whether other distribution areas simultaneously experience common causes such as abnormal bus losses, concentrated outbreaks of communication anomalies, or sudden changes in the external environment. If no obvious common-cause interference is found, and meters from the same batch show the same error trend in multiple independent distribution areas, then this type of change is more likely to be caused by synchronous drift due to component inconsistencies in the meter batch itself, firmware defects in the same version, or manufacturing deviations in the same process. Finally, the system marks the current target meter or related meters from the same batch within the current distribution area as synchronous drift type and records the criteria for judgment, including the number of distribution areas participating in the comparison, the number of samples from the same batch, the trend consistency ratio, the average rate of change, and the corresponding time window. Through this judgment method, local drift caused by individual environmental factors and group drift caused by equipment batch characteristics can be effectively distinguished, thereby providing a basis for subsequent unified correction or batch processing, improving the accuracy of analysis and processing efficiency.

[0039] Based on the identified drift type, adaptive verification processing is performed to determine the verification drift type and error magnitude, and a hierarchical correction strategy is executed. The hierarchical correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

[0040] Specifically, after identifying the drift type, an adaptive verification phase is initiated for different types of metering drift. During this phase, based on the identified drift type, a matching verification method is selected to specifically review the meter's anomalies, further confirming the accuracy of the drift type and quantifying its error magnitude. During verification, the system combines current meter operating data, historical data, and necessary comparative reference data to recalculate and analyze the error. For example, by comparing metering differences between different time periods, operating conditions, or similar meters, the system determines whether the error is stable and its range of variation, thus obtaining a more reliable error assessment result. Simultaneously, the verification process eliminates misjudgments, ensuring subsequent processing is based on high-confidence judgments. After completing verification and determining the final drift type and error magnitude, the system executes a tiered correction strategy according to preset rules. For situations with small errors, clear patterns of change, and the ability to be corrected through parameter adjustments, corresponding correction coefficients can be generated and remotely distributed to the electricity meter or related systems for soft correction, requiring no on-site intervention. For situations with large errors, sudden anomalies, or potential equipment malfunctions, on-site maintenance work orders are automatically generated, arranging for personnel to conduct on-site repairs or replace the equipment. For situations with abnormal electricity consumption characteristics that may involve electricity theft, an anti-theft audit task is triggered, initiating a special investigation process. By organically combining drift identification, verification, and tiered correction, closed-loop management from anomaly detection to handling execution can be achieved, ensuring processing accuracy while improving operational efficiency and reducing labor costs.

[0041] In some embodiments, based on the identified drift type, adaptive verification processing is performed to determine the verification drift type and error magnitude, and a graded correction strategy is executed, including: When the drift is identified as a light-load nonlinear drift, active probe verification is performed. During a preset low-load period, a coded characteristic signal of known energy is injected into the branch containing the candidate energy meter, and the deviation between the meter reading and the injected value is compared. When the drift is identified as a synchronous drift, a horizontal comparison verification within the same branch is performed. Healthy energy meters with similar load characteristics are selected from the same branch as a reference group, and the deviation of the metering curves of the candidate energy meter and the reference group is compared. When the drift is identified as a temperature-coupled drift, a seasonal term removal verification is performed. The seasonal term is removed from the line loss deviation, and the adjusted deviation is calculated to verify the correlation between drift and temperature. When the drift is identified as a component aging drift, a long-term trend verification is performed to confirm that the trend term persists and its rate of change is stable. When the drift is identified as a communication failure drift, a re-acquisition verification and time synchronization verification are performed. The communication quality tags before and after the abrupt change are read; if the communication quality is normal, on-site verification is triggered. When the drift is identified as an electricity theft-related drift, a behavioral analysis verification is performed to analyze the correlation between the error abrupt change time and the user's electricity consumption period. Based on the drift type and error magnitude determined by the verification, the corresponding correction strategy is executed.

[0042] Specifically, for different identified drift types, the system employs a differentiated adaptive verification process for further confirmation. After verification, the error magnitude is determined and corresponding correction strategies are implemented. When identified as a light-load nonlinear drift, the system injects a characteristic signal with a known energy value, specific frequency, and time code into the branch where the candidate energy meter is located during a preset low-load period. The amplitude of this signal is controlled within a range that does not affect the user's normal electricity consumption. Simultaneously, the injected energy reference value is recorded, and the metering value of the candidate energy meter during this period is collected and compared with the injected reference value to calculate the deviation, thereby verifying whether there is a significant nonlinear error under low-current conditions. When identified as a synchronous drift, the system selects several energy meters with similar load characteristics, stable historical performance, and no abnormal markings within the same branch range as a reference group. By comparing the candidate energy meters with the reference value, the system further verifies whether there is a significant nonlinear error under low-current conditions. The system selects the metering curves of the energy meter and the reference group within the same time window, including power curves or energy consumption curves, and calculates the curve difference index, such as mean square error or normalized deviation. If the candidate meter and the reference group are offset as a whole and in the same direction, it verifies that it is a batch consistency drift. When it is identified as a temperature-coupled drift, the system removes the seasonal terms obtained from the time series decomposition from the original line loss deviation or error series to obtain a deseasoned adjustment deviation series, and recalculates the correlation between this series and the ambient temperature. If the deviation is significantly reduced after removing the seasonal terms and the relationship with temperature changes is significantly weakened, it verifies that the drift is mainly caused by temperature coupling. When it is identified as a component... When aging-related drift occurs, the system expands the analysis time window, such as to several months or half a year, to conduct long-term tracking analysis of the trend term, verifying whether it persists and whether the rate of change is basically stable, while eliminating short-term fluctuations. This confirms that the drift is caused by gradual aging and estimates the current error level accordingly. When communication failure-related drift is identified, the system performs a re-acquisition operation, that is, actively reads the meter data multiple times in a short period of time, while performing time synchronization verification to check whether the meter clock is consistent with the master station clock, and reads the communication quality tags before and after the error mutation. If the anomaly is found to be caused by a communication problem, it is marked as a communication failure; otherwise, if the communication quality is normal... If an anomaly persists under the current circumstances, an on-site verification process is triggered. When an anomaly is identified as a combination of electricity theft and other types, the system performs correlation analysis between the moment of error mutation and the user's electricity consumption behavior. It comprehensively analyzes load curve anomalies, reverse power records, electricity consumption patterns, open cover or power outage events, and historical inspection records to determine whether there is human intervention or abnormal power consumption behavior. After completing the above verifications for different types of users, the system uniformly outputs the final verification drift type and corresponding error magnitude, including absolute and relative errors. It also executes the corresponding correction strategy in conjunction with a preset hierarchical strategy, thereby achieving closed-loop management from identification and verification to handling, and improving the overall accuracy and efficiency of processing.

[0043] In some embodiments, a corresponding correction strategy is executed based on the verified drift type and error magnitude, including: When the drift is verified as component aging type, temperature coupling type, light load nonlinear type, or synchronous drift type, and the absolute value of the error is within the first preset interval and the confidence level is greater than the preset confidence threshold, a corresponding correction coefficient is generated and remote soft correction is performed; when the drift is verified as sudden failure type or the absolute value of the error is greater than the second preset threshold, an on-site maintenance work order is generated; when the drift is verified as electricity theft superimposed type, an anti-electricity theft inspection task is triggered; when the drift is verified as communication failure type, a communication maintenance work order is triggered; wherein, the remote soft correction includes: version management of the correction coefficient, and recalculating the adjusted line loss deviation within a preset evaluation window after performing remote soft correction; when the adjusted line loss deviation does not converge or exceeds the rollback threshold, it is automatically rolled back to the previous correction coefficient version.

[0044] Specifically, after completing the drift type verification, the system first obtains the final verified drift type, absolute error value, error direction, and confidence index for each candidate energy meter, and matches them with pre-set grading rules. The confidence index is a weighted average of data quality, the proportion of consecutive deviations exceeding the threshold, and the consistency of verification results. Data quality is calculated as: real-time data acquisition percentage × (1 - supplementary data percentage) × (1 - average communication delay / maximum allowable communication delay). When the verification results indicate that the meter belongs to component aging drift, temperature coupling drift, light-load nonlinear drift, or synchronous drift, and its absolute error value falls within the first preset range and the confidence level is higher than the preset confidence threshold, the system automatically enters the remote soft correction process. Specifically, corresponding correction parameters are generated according to different drift types. These correction parameters are used to proportionally correct the current metering result of the energy meter, and their calculation basis is the error magnitude obtained during the verification stage, typically 1 plus the reciprocal of the relative error. Subsequently, the master station sends the correction parameters to the target energy meter or concentrator-side metering compensation unit for execution. During the distribution process, the system manages each set of correction parameters using versioning, generating a unique version identifier, recording its effective time, applicable objects, generation basis, and associated drift type, and establishing a mapping relationship between the new version and the previous version to support rollback operations. After remote soft calibration is executed, the system continuously monitors the changes in line loss deviation corresponding to the meter within a preset evaluation window, recalculates the deviation between the corrected measured line loss rate and the theoretical line loss benchmark value, and determines whether the deviation gradually converges to the normal range. If the deviation is observed to decrease significantly and stabilize within the allowable range within the evaluation window, the soft calibration is deemed effective, and the current correction version is solidified as the effective version for continued use. If the deviation does not converge within the evaluation window, but instead expands or exceeds the preset rollback threshold, the current correction parameters are deemed incompatible or misjudged, and the system automatically executes the rollback mechanism, restoring the correction parameters to the previous version, while recording this rollback event and marking the meter as an object requiring further manual review.

[0045] When the verification result indicates that the meter is experiencing a sudden fault-type drift, or although it does not belong to this type, its absolute error value exceeds the second preset threshold, the system will not perform soft correction but will automatically generate a field maintenance work order. This field maintenance work order must include at least the meter identification, the transformer area and meter box to which it belongs, the time of the anomaly, the verified drift type, the error amplitude, recommended handling measures, and priority information, and will be pushed to the maintenance system to arrange for on-site personnel to carry out inspection, replacement, or wiring verification. When the verification result is an electricity theft-related type, the system will directly classify the anomaly into the audit process, trigger an anti-electricity theft audit task, package the relevant evidence data and push it to the audit platform for further verification and evidence collection by dedicated personnel. When the verification result is a communication fault type, the system will generate a communication maintenance work order, instructing the meter or its communication link to be investigated and repaired. Through the above process, differentiated handling based on different drift types, error magnitudes, and confidence levels is achieved. This ensures measurement accuracy while avoiding unnecessary on-site maintenance costs. Furthermore, version management and rollback mechanisms enhance the security and reliability of remote soft calibration, thereby forming a complete closed-loop calibration and maintenance decision-making system that improves overall maintenance efficiency and system stability.

[0046] In summary, the metering drift correction method for energy meters based on transformer substation line loss provided by this invention has the following technical effects: A four-level hierarchical topology structure of transformer area-branch-meter box-meter is constructed, and a communication quality tag is attached to each data collection line. Based on this hierarchical topology, the measured line loss rate is calculated layer by layer, and a theoretical line loss benchmark value is calculated using a graph neural network embedded with a physical model of line loss. The deviation between the measured line loss rate and the theoretical line loss benchmark value is calculated, and the communication quality tag of the corresponding data is read to identify suspicious drift candidate nodes. For the suspicious drift candidate meters corresponding to the suspicious drift candidate nodes, a time series decomposition of historical error change trajectories is performed to identify the drift type. Based on the identified drift type, adaptive verification processing is performed to determine the verified drift type and error magnitude, and a hierarchical correction strategy is executed. The hierarchical correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks. This achieves the technical effect of improving the accuracy of line loss analysis by constructing a refined topology and graph neural network, and accurately identifying drift types by combining communication quality tags and time series decomposition, thereby improving the reliability of electricity meter drift handling and the efficiency of operation and maintenance.

[0047] Example 2, as Figure 2 This is a schematic diagram of the energy meter drift correction system based on transformer substation line loss according to the present invention. For example, Figure 1 The flowchart of the energy meter drift correction method based on transformer substation line loss of the present invention can be seen as follows: Figure 2 The structure shown is implemented.

[0048] Based on the same concept as the electricity meter drift correction method based on transformer area line loss in the above embodiments, the present invention also provides an electricity meter drift correction system based on transformer area line loss, comprising: Topology Construction Module 11: Constructs a four-level hierarchical topology structure of transformer area-branch-meter box-meter, and adds a communication quality label to each collected data; Theoretical Line Loss Calculation Module 12: Calculates the measured line loss rate based on the hierarchical topology structure, and calculates the theoretical line loss benchmark value using a graph neural network embedded with a physical model of line loss; Line Loss Deviation Identification Module 13: Calculates the deviation between the measured line loss rate and the theoretical line loss benchmark value, and reads the communication quality label of the corresponding data to identify suspicious drift candidate nodes; Time Series Decomposition Module 14: Performs time series decomposition of historical error change trajectories on the suspicious drift candidate meters corresponding to the suspicious drift candidate nodes to identify the drift type; Graded Correction Module 15: Based on the identified drift type, performs adaptive verification processing to determine the verified drift type and error magnitude, and executes a graded correction strategy, which includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

[0049] In some embodiments, the topology building module 11 includes: A characteristic current pulse with preset amplitude, frequency, and time encoding is injected into the transformer area. Based on the response characteristics of each sub-meter to the characteristic current pulse, the affiliation between the meter and the branch / meter box is automatically identified, generating a four-level hierarchical topology structure of transformer area-branch-meter box-meter. The power consumption data, voltage and current curves, and operating status data of each sub-meter are collected, and a communication quality tag is attached to each collected data. The communication quality tag includes at least the collection time, communication delay, and recall flag. The recall flag is generated as follows: when the meter data cannot be successfully read within the collection period, and the data is supplemented from the cache or historical data, the recall flag of this data is set to true, and the number of recalls is recorded. When the number of consecutive recalls exceeds a preset recall threshold, the meter is marked as a communication abnormal meter.

[0050] In some embodiments, the theoretical line loss calculation module 12 includes: Based on the hierarchical topology, the ratio of the difference between the input power of the parent node and the output power of the child node to the input power of the parent node is calculated to obtain the measured line loss rate of each node. The hierarchical topology is mapped to a graph structure, where nodes correspond to main meters, branch nodes, meter box nodes, and energy meter nodes, and edges correspond to parent-child connection relationships. Using node features and edge features as input, a graph neural network model is used for forward propagation to output the theoretical line loss baseline value of each node. The node features include at least node type, historical power, real-time voltage, real-time current, power factor, and load fluctuation characteristics. The edge features include at least line length, conductor type, equivalent impedance, real-time current, ambient temperature, and topology level. The graph neural network model has physical constraints on line loss, which include at least line ohmic loss constraints and node power conservation constraints.

[0051] In some embodiments, the line loss deviation identification module 13 includes: If the deviation between the measured line loss rate and the theoretical line loss benchmark value exceeds a preset range, and the communication quality label shows that the data is collected in real time and the communication latency does not exceed the standard, then the corresponding node will be marked as a suspected drift candidate node; if the communication quality label shows that the data is supplementary data or the communication latency exceeds the standard, then the corresponding node will be marked as a communication suspicious node, and will not enter the drift type identification process for the time being, and will wait for at least one collection cycle to re-evaluate the drift suspicion based on the newly collected data.

[0052] In some embodiments, the time series decomposition module 14 includes: The historical error change trajectory time series of the identified suspected drift candidate meters is decomposed to determine the trend term, seasonal term, and random term. If the trend term is monotonic and the absolute value of the rate of change is less than a first rate of change threshold, it is determined to be component aging type drift. If the correlation coefficient between the seasonal term and the ambient temperature data is greater than a first correlation coefficient threshold, it is determined to be temperature-coupled type drift. If the random term has a step change and the step amplitude is greater than a preset amplitude threshold, it is determined to be sudden failure type drift. If the error of the suspected drift candidate meter increases significantly under low current conditions and has a nonlinear relationship with the current magnitude, it is determined to be light-load nonlinear type drift.

[0053] In some embodiments, the time series decomposition module 14 includes: Establish a cross-regional meter batch database to record the error statistics of meters in the same batch across multiple regions. When the error trend terms of multiple meters in the same region are consistent in direction and have similar rates of change, and the bus loss rate of that region does not exceed the abnormal threshold, the cross-regional meter batch database is called for horizontal comparison. If meters in the same batch show the same error trend in other regions, it is determined to be a synchronous drift type drift.

[0054] In some embodiments, the graded correction module 15 includes: When the drift is identified as a light-load nonlinear drift, active probe verification is performed. During a preset low-load period, a coded characteristic signal of known energy is injected into the branch containing the candidate energy meter, and the deviation between the meter reading and the injected value is compared. When the drift is identified as a synchronous drift, a horizontal comparison verification within the same branch is performed. Healthy energy meters with similar load characteristics are selected from the same branch as a reference group, and the deviation of the metering curves of the candidate energy meter and the reference group is compared. When the drift is identified as a temperature-coupled drift, a seasonal term removal verification is performed. The seasonal term is removed from the line loss deviation, and the adjusted deviation is calculated to verify the correlation between drift and temperature. When the drift is identified as a component aging drift, a long-term trend verification is performed to confirm that the trend term persists and its rate of change is stable. When the drift is identified as a communication failure drift, a re-acquisition verification and time synchronization verification are performed. The communication quality tags before and after the abrupt change are read; if the communication quality is normal, on-site verification is triggered. When the drift is identified as an electricity theft-related drift, a behavioral analysis verification is performed to analyze the correlation between the error abrupt change time and the user's electricity consumption period. Based on the drift type and error magnitude determined by the verification, the corresponding correction strategy is executed.

[0055] In some embodiments, the graded correction module 15 includes: When the drift is verified as component aging type, temperature coupling type, light load nonlinear type, or synchronous drift type, and the absolute value of the error is within the first preset interval and the confidence level is greater than the preset confidence threshold, a corresponding correction coefficient is generated and remote soft correction is performed; when the drift is verified as sudden failure type or the absolute value of the error is greater than the second preset threshold, an on-site maintenance work order is generated; when the drift is verified as electricity theft superimposed type, an anti-electricity theft inspection task is triggered; when the drift is verified as communication failure type, a communication maintenance work order is triggered; wherein, the remote soft correction includes: version management of the correction coefficient, and recalculating the adjusted line loss deviation within a preset evaluation window after performing remote soft correction; when the adjusted line loss deviation does not converge or exceeds the rollback threshold, it is automatically rolled back to the previous correction coefficient version.

[0056] In embodiment three, the present invention also provides a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the electricity meter metering drift correction method based on transformer area line loss in the embodiments of the present invention, thereby realizing the above-mentioned electricity meter metering drift correction method based on transformer area line loss.

[0057] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. A method for correcting meter drift in electricity meters based on distribution transformer line loss, characterized in that, include: Construct a four-level hierarchical topology structure of transformer area-branch-meter box-meter, and attach a communication quality label to each piece of collected data; The measured line loss rate is calculated in layers based on the hierarchical topology, and the theoretical line loss benchmark value is calculated using a graph neural network embedded with the physical model of line loss. Calculate the deviation between the measured line loss rate and the theoretical line loss benchmark value, and read the communication quality label of the corresponding data to identify suspected drift candidate nodes; For the suspected drift candidate meters corresponding to the suspected drift candidate nodes, perform time series decomposition of historical error change trajectories to identify the drift type; Based on the identified drift type, adaptive verification processing is performed to determine the verification drift type and error magnitude, and a hierarchical correction strategy is executed. The hierarchical correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

2. The method for correcting meter drift based on transformer substation line loss according to claim 1, characterized in that, A four-level hierarchical topology of transformer area-branch-meter box-meter is constructed, and a communication quality label is attached to each piece of collected data, including: Inject a characteristic current pulse with a preset amplitude, frequency, and time code into the transformer area; Based on the response characteristics of each sub-meter to the characteristic current pulse, the affiliation between the meter and the branch and meter box is automatically identified, and a four-level hierarchical topology structure of transformer area-branch-meter box-meter is generated. Collect power data, voltage and current curves, and operating status data from each sub-meter, and attach a communication quality tag to each piece of collected data. The communication quality tag includes at least the collection time, communication delay, and recall identifier. The generation rule for the supplementary recall flag is as follows: when the meter data cannot be successfully read within the collection period, and data is supplemented from the cache or historical data, the supplementary recall flag for this data is set to true, and the number of supplementary recalls is recorded. When the number of consecutive recall attempts exceeds the preset recall attempt threshold, the meter will be marked as a communication malfunction meter.

3. The method for correcting meter drift based on transformer substation line loss according to claim 2, characterized in that, Based on the hierarchical topology, the measured line loss rate is calculated in layers, and the theoretical line loss benchmark value is calculated using a graph neural network embedded with the physical model of line loss, including: Based on the hierarchical topology, the ratio of the difference between the input power of the parent node and the output power of the child node to the input power of the parent node is calculated to obtain the measured line loss rate of each node. The hierarchical topology is mapped to a graph structure, where nodes correspond to master table, branch nodes, meter box nodes and energy meter nodes, and edges correspond to parent-child connection relationships. Using node features and edge features as input, a graph neural network model is used for forward propagation to output the theoretical line loss baseline value for each node; The node features include at least node type, historical power consumption, real-time voltage, real-time current, power factor, and load fluctuation characteristics. The edge features include at least line length, conductor type, equivalent impedance, real-time current, ambient temperature, and topology level. The graph neural network model has physical constraints on line losses, which include at least line ohmic loss constraints and node power conservation constraints.

4. The method for correcting meter drift based on transformer substation line loss according to claim 1, characterized in that, Calculate the deviation between the measured line loss rate and the theoretical line loss benchmark value, and read the communication quality tags of the corresponding data to identify suspected drift candidate nodes, including: If the deviation between the measured line loss rate and the theoretical line loss benchmark value exceeds the preset range, and the communication quality label shows that the data is collected in real time and the communication delay does not exceed the standard, then the corresponding node will be marked as a suspected drift candidate node. If the communication quality label shows data that is missing or communication latency exceeds the standard, the corresponding node will be marked as a communication suspicious node, and the drift type identification process will not be entered for the time being. The drift suspicion assessment will be carried out again based on the newly collected data after at least one collection cycle.

5. The method for correcting meter drift based on transformer substation line loss according to claim 1, characterized in that, For the suspected drift candidate meters corresponding to the suspected drift candidate nodes, perform historical error change trajectory time series decomposition to identify the drift type, including: The historical error change trajectory time series of the identified suspected drift candidate meters is decomposed to determine the trend term, seasonal term, and random term; If the trend term is monotonic and the absolute value of the rate of change is less than the first rate of change threshold, it is determined to be component aging drift. If the correlation coefficient between the seasonal term and the ambient temperature data is greater than the first correlation coefficient threshold, it is determined to be a temperature-coupled drift. If the random item has a step change and the step amplitude is greater than a preset amplitude threshold, it is determined to be a sudden failure type drift. If the suspected drift candidate meter shows a significant increase in error under low current conditions and exhibits a non-linear relationship with the current magnitude, it is determined to be a light-load non-linear drift type.

6. The method for correcting meter drift based on transformer substation line loss according to claim 5, characterized in that, Identifying drift types also includes: Establish a cross-regional electricity meter batch database to record the error statistics of the same batch of electricity meters in multiple regions; When the error trend terms of multiple meters in the same distribution area are in the same direction and have similar rates of change, and the bus loss rate of that distribution area does not exceed the abnormal threshold, the cross-distribution meter batch database is called for horizontal comparison. If meters from the same batch exhibit the same error trend in other distribution areas, it is determined to be a synchronous drift type of drift.

7. The method for correcting meter drift based on transformer substation line loss according to claim 6, characterized in that, Based on the identified drift type, adaptive verification processing is performed to determine the verification drift type and error magnitude, and a hierarchical correction strategy is implemented, including: When a light-load nonlinear drift is identified, active probe verification is performed. During a preset low-load period, a coded feature signal of known energy is injected into the branch where the candidate energy meter is located, and the deviation between the energy meter reading and the injected value is compared. When the drift is identified as synchronous drift, a horizontal comparison verification is performed within the same branch. Healthy energy meters with similar load characteristics are selected from the same branch as a reference group, and the metering curve deviation between the candidate energy meters and the reference group is compared. When the drift is identified as temperature-coupled, seasonal term stripping verification is performed. After removing the seasonal term from the line loss deviation, the adjusted deviation is calculated to verify the correlation between drift and temperature. When the component aging drift is identified, long-term trend verification is performed to confirm that the trend term persists and the rate of change is stable. When a communication failure-type drift is identified, re-acquisition verification and time synchronization verification are performed. The communication quality tags before and after the abrupt change are read. If the communication quality is normal, on-site verification is triggered. When identified as a type of electricity theft, behavioral analysis and verification are performed to analyze the correlation between the time of error mutation and the user's electricity consumption period; Based on the verified drift type and error magnitude, the corresponding correction strategy is executed.

8. The method for correcting meter drift based on transformer substation line loss according to claim 7, characterized in that, Based on the verified drift type and error magnitude, implement the corresponding correction strategy, including: When the drift is verified to be component aging type, temperature coupling type, light load nonlinear type, or synchronous drift type, and the absolute value of the error is within the first preset range and the confidence level is greater than the preset confidence level threshold, the corresponding correction coefficient is generated and remote soft correction is performed. When the verification shows a sudden fault type drift or the absolute value of the error is greater than the second preset threshold, an on-site maintenance work order is generated. When verified as a case of electricity theft with additional charges, an anti-electricity theft investigation task is triggered. When the problem is verified as a communication failure, a communication maintenance work order is triggered. The remote soft correction includes: versioning the correction coefficients and recalculating the adjusted line loss deviation within a preset evaluation window after performing the remote soft correction; If the adjusted line loss deviation fails to converge or exceeds the rollback threshold, it will automatically roll back to the previous correction factor version.

9. A metering drift correction system for electricity meters based on distribution transformer line loss, characterized in that, The method for correcting meter drift based on transformer substation line loss as described in any one of claims 1-8 includes: Topology building module: Constructs a four-level hierarchical topology structure of transformer area-branch-meter box-meter, and adds a communication quality label to each piece of collected data; Theoretical line loss calculation module: Based on the hierarchical topology, the measured line loss rate is calculated in layers, and the theoretical line loss benchmark value is calculated using a graph neural network embedded with the physical model of line loss; Line loss deviation identification module: Calculates the deviation between the measured line loss rate and the theoretical line loss benchmark value, and reads the communication quality label of the corresponding data to identify suspected drift candidate nodes; Time series decomposition module: Performs time series decomposition of historical error change trajectories on the suspected drift candidate meters corresponding to the suspected drift candidate nodes to identify the drift type; The graded correction module performs adaptive verification processing based on the identified drift type, determines the verification drift type and error magnitude, and executes a graded correction strategy. The graded correction strategy includes remotely issuing correction coefficients for soft correction, generating on-site maintenance work orders, or triggering anti-electricity theft inspection tasks.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the metering drift correction method for electricity meters based on transformer substation line loss as described in any one of claims 1 to 8.

Images