Intelligent electric meter data anomaly detection and quality evaluation optimization method and system

Abstract

The invention provides an intelligent electric meter data anomaly detection and quality evaluation optimization method and system, and relates to the technical field of intelligent power grids, and the method comprises the steps: obtaining the measurement data of electric energy metering equipment, and constructing a graph model based on a topological relation; extracting multi-order correlation features through a graph structure propagation mechanism; inputting the features and the measured data into an anomaly detection model to calculate a deviation degree, and judging anomaly if the deviation degree exceeds a threshold value; and calculating a data quality evaluation index according to the abnormal frequency and the duration. According to the method, network topology constraints are fully utilized, the anomaly detection accuracy can be improved, and accurate evaluation of ammeter data quality is realized.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and in particular to a method and system for detecting and optimizing data anomalies in smart meters. Background Technology

[0002] With the rapid development of smart grids, smart meters have been widely deployed in distribution networks for measuring and billing electricity consumption. The large amount of data generated by smart meters is of great value for grid operation monitoring, load analysis, and electricity demand forecasting. However, due to various factors such as equipment failure, communication interference, and environmental interference, smart meter data often becomes abnormal, affecting the accuracy and reliability of the data, and consequently impacting the decision-making and operational efficiency of power companies.

[0003] Traditional methods for detecting anomalies in smart meter data primarily rely on statistical analysis or machine learning algorithms, identifying abnormal patterns by analyzing historical data from individual meters. These methods typically only consider data variation patterns over time, neglecting the spatial correlations between meters. With the increasing complexity of power grids and the explosive growth of data volume, traditional methods face problems of inefficiency and insufficient accuracy. Summary of the Invention

[0004] This invention provides a method and system for detecting and optimizing abnormal data in smart meters, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for detecting and optimizing the quality of smart meter data anomalies, comprising: A time-series measurement data of multiple energy metering devices is acquired. Based on the topological connection relationship and physical constraint relationship of the power distribution network, the multiple energy metering devices are mapped as nodes. The measurement data is used as node attributes, and the topological connection relationship and physical constraint relationship are used as edge relationships to construct an energy metering device graph model. In the power metering equipment graph model, multi-level association features between the target power metering equipment node and its adjacent nodes in the time and space dimensions are extracted through the graph structure propagation mechanism to obtain the association feature representation; The correlation feature representation and the measurement data of the target power metering device node are input into the anomaly detection model. The anomaly detection model calculates the degree of deviation of the measurement data under the constraint of the correlation feature representation. When the degree of deviation exceeds a preset deviation threshold, it is determined that the measurement data is abnormal, and the anomaly detection result is obtained. Based on the anomaly detection results, the frequency and duration of anomalies occurring in the target energy metering device node within a preset time window are statistically analyzed. Based on the frequency and duration of anomalies, the data quality evaluation index of the target energy metering device node is calculated.

[0006] Based on the topological connections and physical constraints of the power distribution network, node mapping is performed on the multiple power metering devices. The measurement data is used as node attributes, and the topological connections and physical constraints are used as edge relationships. A graph model of the power metering devices is constructed, including: The topological connections of the power distribution network are analyzed to determine the upstream and downstream connections and branch affiliations among the multiple power metering devices, thus obtaining the topological connection structure. Based on the topological connection structure, the physical constraint relationship between the multiple power metering devices is extracted to obtain the physical constraint structure; The multiple energy metering devices are mapped as graph nodes, and the voltage measurement data, current measurement data, and power measurement data in the measurement data are used as the node attributes of the corresponding graph nodes to construct a node set containing the graph nodes corresponding to the multiple energy metering devices; Based on the topological connection structure, topological connection edges are established between graph nodes with upstream and downstream connections. Based on the physical constraint structure, physical constraint edges are established between graph nodes with physical law constraints, resulting in an edge set. The power metering equipment graph model is generated based on the set of nodes and the set of edges. The topological connection edges and physical constraint edges in the power metering equipment graph model jointly represent the association relationship between the multiple power metering devices.

[0007] In the power metering equipment graph model, multi-level association features between the target power metering equipment node and its adjacent nodes in the time and space dimensions are extracted through the graph structure propagation mechanism, resulting in association feature representations including: Centered on the target energy metering device node, extend outwards along the topological connection edge and the physical constraint edge by a preset propagation order to determine adjacent nodes that have a direct or indirect connection relationship with the target energy metering device node, thus obtaining a set of adjacent nodes; The change pattern of the measurement data of the target energy metering device node in the time series is obtained, the change pattern of the measurement data of each neighboring node in the neighboring node set in the time series is obtained, and the time synchronization and temporal correlation between the change pattern of the measurement data of the target energy metering device node and the change pattern of the measurement data of each neighboring node are calculated to obtain the time dimension correlation features; Based on the topological connection edges and physical constraint edges between the target energy metering device node and the set of adjacent nodes in the energy metering device graph model, the node attributes of each adjacent node in the set of adjacent nodes are propagated to the target energy metering device node along the edge relationships through a graph structure propagation mechanism. The propagated node attributes are weighted and aggregated according to the propagation path length and edge relationship type to obtain spatial dimension association features. The temporal dimension association features and the spatial dimension association features are fused to obtain the association feature representation.

[0008] The anomaly detection model calculates the degree of deviation of the measurement data under the constraints of the associated feature representation. When the degree of deviation exceeds a preset deviation threshold, it is determined that the measurement data is abnormal, and the anomaly detection result includes: The associated feature representation is input into the anomaly detection model, and the anomaly detection model performs prediction calculations on the measurement data of the target power metering equipment node based on the associated feature representation to obtain the predicted measurement data; Calculate the difference metric between the actual measurement data and the predicted measurement data of the target energy metering device node. The difference metric characterizes the degree to which the actual measurement data deviates from the normal value range constrained by the correlation feature representation, and obtain the degree of deviation. The deviation degree is compared with the preset deviation threshold. When the deviation degree exceeds the preset deviation threshold, it is determined that the measurement data of the target power metering device node is abnormal and an abnormality mark is generated. When the deviation degree does not exceed the preset deviation threshold, it is determined that the measurement data of the target power metering device node is normal and a normality mark is generated, thus obtaining the abnormality detection result.

[0009] The anomaly detection model predicts the measurement data of the target power metering device node based on the associated feature representation, resulting in predicted measurement data including: Based on the historical measurement data of the target energy metering device node and the time dimension correlation features in the correlation feature representation, the initial prediction measurement data of the target energy metering device node is generated through the initial prediction module of the anomaly detection model; The difference between the initial predicted measurement data and the historical measurement data of the target power metering equipment node is calculated, and the residual information that the initial predicted measurement data fails to represent is extracted to obtain the prediction residual characteristics; The predicted residual features and the spatial dimension correlation features in the correlation feature representation are input into the graph convolution correction module of the anomaly detection model. The predicted residual features are then propagated along the topological connection edges and physical constraint edges in the power metering equipment graph model through the graph convolution correction module. The predicted residual features are corrected based on the spatial constraint relationship of adjacent nodes to obtain the corrected residual value. The initial predicted measurement data is superimposed with the corrected residual value to obtain the predicted measurement data.

[0010] The data quality assessment indicators for the target energy metering equipment node are calculated based on the frequency of anomalies and the duration of anomalies, including: The frequency of anomalies occurring at the target energy metering equipment nodes within a preset evaluation time window is statistically analyzed, and the ratio of the anomaly frequency to the total number of samples within the preset evaluation time window is calculated to obtain the anomaly frequency percentage. The abnormal duration of all abnormal events at the target energy metering device node within the preset evaluation time window is statistically analyzed, and the ratio of the abnormal duration to the total duration of the preset evaluation time window is calculated to obtain the abnormal duration percentage. The frequency impact weight is determined based on the degree of influence of the abnormality frequency ratio on data quality, and the duration impact weight is determined based on the degree of influence of the abnormality duration ratio on data quality. The sum of the frequency impact weight and the duration impact weight is a normalized value. The frequency quality loss value is obtained by weighting the abnormal frequency ratio with the frequency impact weight, and the duration quality loss value is obtained by weighting the abnormal duration ratio with the duration impact weight. The frequency quality loss value and the duration quality loss value are aggregated to obtain the total quality loss value. The data quality assessment index of the target power metering equipment node is calculated based on the total quality loss value.

[0011] A second aspect of the present invention provides a smart meter data anomaly detection and quality assessment optimization system, comprising: The first unit is used to acquire measurement data of multiple energy metering devices over time, perform node mapping on the multiple energy metering devices based on the topological connection relationship and physical constraint relationship of the power distribution network, use the measurement data as node attributes, and use the topological connection relationship and physical constraint relationship as edge relationship to construct an energy metering device graph model; The second unit is used to extract multi-level association features between the target energy metering device node and its adjacent nodes in the time and space dimensions through a graph structure propagation mechanism in the energy metering device graph model, and obtain the association feature representation. The third unit is used to input the correlation feature representation and the measurement data of the target power metering device node into the anomaly detection model, calculate the degree of deviation of the measurement data under the constraint of the correlation feature representation through the anomaly detection model, and determine that the measurement data is abnormal when the degree of deviation exceeds a preset deviation threshold, thereby obtaining the anomaly detection result; The fourth unit is used to count the frequency and duration of anomalies of the target power metering device node within a preset time window based on the anomaly detection results, and to calculate the data quality evaluation index of the target power metering device node based on the frequency and duration of anomalies.

[0012] A third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0013] Fourth aspect of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] The beneficial effects of this application are as follows: By constructing a graph model of electricity metering equipment and organically combining measurement data with network structure using the topological connections and physical constraints of the power distribution network, the inherent correlation between electricity metering equipment is fully explored, thereby improving the accuracy and reliability of anomaly detection.

[0015] This method innovatively applies the graph structure propagation mechanism and extracts multi-level association features in both time and space dimensions. Compared with traditional methods, it can capture data anomaly patterns more comprehensively, reduce false alarm rate, and enhance the model's adaptability to complex scenarios.

[0016] This method can not only detect abnormal data, but also construct a data quality assessment index system by statistically analyzing the frequency and duration of abnormal occurrences. This provides a quantitative basis for the operation and maintenance management of power metering equipment and the evaluation of data reliability, and helps to improve the overall monitoring level and operating efficiency of the power distribution network. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for detecting and optimizing abnormal data in smart meters according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating the smart meter data anomaly detection and quality assessment optimization method according to an embodiment of the present invention. Figure 1 As shown, the method includes: A time-series measurement data of multiple energy metering devices is acquired. Based on the topological connection relationship and physical constraint relationship of the power distribution network, the multiple energy metering devices are mapped as nodes. The measurement data is used as node attributes, and the topological connection relationship and physical constraint relationship are used as edge relationships to construct an energy metering device graph model. In the power metering equipment graph model, multi-level association features between the target power metering equipment node and its adjacent nodes in the time and space dimensions are extracted through the graph structure propagation mechanism to obtain the association feature representation; The correlation feature representation and the measurement data of the target power metering device node are input into the anomaly detection model. The anomaly detection model calculates the degree of deviation of the measurement data under the constraint of the correlation feature representation. When the degree of deviation exceeds a preset deviation threshold, it is determined that the measurement data is abnormal, and the anomaly detection result is obtained. Based on the anomaly detection results, the frequency and duration of anomalies occurring in the target energy metering device node within a preset time window are statistically analyzed. Based on the frequency and duration of anomalies, the data quality evaluation index of the target energy metering device node is calculated.

[0021] In one optional implementation, the plurality of power metering devices are mapped to nodes based on the topological connections and physical constraints of the power distribution network. The measurement data is used as node attributes, and the topological connections and physical constraints are used as edge relationships. The power metering device graph model is constructed by: The topological connections of the power distribution network are analyzed to determine the upstream and downstream connections and branch affiliations among the multiple power metering devices, thus obtaining the topological connection structure. Based on the topological connection structure, the physical constraint relationship between the multiple power metering devices is extracted to obtain the physical constraint structure; The multiple energy metering devices are mapped as graph nodes, and the voltage measurement data, current measurement data, and power measurement data in the measurement data are used as the node attributes of the corresponding graph nodes to construct a node set containing the graph nodes corresponding to the multiple energy metering devices; Based on the topological connection structure, topological connection edges are established between graph nodes with upstream and downstream connections. Based on the physical constraint structure, physical constraint edges are established between graph nodes with physical law constraints, resulting in an edge set. The power metering equipment graph model is generated based on the set of nodes and the set of edges. The topological connection edges and physical constraint edges in the power metering equipment graph model jointly represent the association relationship between the multiple power metering devices.

[0022] In practical applications, the first step is to analyze the topological connections of the power distribution network. This involves analyzing the network's raw data, which typically includes information such as the installation locations and connection methods of electrical equipment. By processing this data, the upstream and downstream connections and branch affiliations between multiple energy metering devices can be determined. For example, consider a power distribution network in a certain area with 10 energy metering devices, numbered M001 to M010. M001 is the master meter, connected downstream to three sub-meters: M002, M003, and M004; M002 is connected downstream to M005 and M006; M003 is connected downstream to M007 and M008; and M004 is connected downstream to M009 and M010. By analyzing this connection information, a clear topological connection structure can be obtained, clarifying the upstream and downstream relationships of each device.

[0023] After obtaining the topology, the physical constraints between multiple power metering devices are extracted based on this structure, resulting in a physical constraint structure. These physical constraints primarily reflect the relationships described by physical laws in power systems, such as Kirchhoff's laws and Ohm's law. Specifically, according to Kirchhoff's current law, the current in an upstream device equals the sum of the currents in all downstream devices; according to Kirchhoff's voltage law, the voltage drop on the same branch must satisfy a specific rule. For example, in the aforementioned distribution network, the current in M001 should equal the sum of the currents in M002, M003, and M004; similarly, the current in M002 should equal the sum of the currents in M005 and M006. These physical constraints constitute the physical constraint structure.

[0024] Next, multiple energy metering devices are mapped to graph nodes, and voltage, current, and power measurement data from the measurement data are used as node attributes of the corresponding graph nodes. For example, for device M001, its measurement data might include: voltage 220V, current 50A, active power 10kW, and reactive power 3kVar. These data are used as attributes of node M001. Similarly, the measurement data of other devices are also used as attributes of their respective nodes. In this way, a node set containing graph nodes corresponding to multiple energy metering devices is constructed.

[0025] Based on the previously obtained topological connection structure, topological connection edges are established between graph nodes with upstream and downstream connections. For example, topological connection edges are established between nodes with upstream and downstream relationships, such as M001 and M002, and M002 and M005. Simultaneously, based on the physical constraint structure, physical constraint edges are established between graph nodes subject to physical law constraints. For example, based on Kirchhoff's current law, physical constraint edges are established between M001 and its downstream nodes M002, M003, and M004, representing the constraint that the current in M001 is equal to the sum of the currents in M002, M003, and M004. In this way, a complete set of edges is obtained.

[0026] Finally, a graph model of electricity metering devices is generated based on the set of nodes and edges. In this model, topological connection edges and physical constraint edges jointly represent the relationships between multiple electricity metering devices. In this way, the graph model of electricity metering devices not only reflects the physical connection relationships between devices, but also includes constraint relationships based on physical laws, enabling the model to more comprehensively describe the structure and characteristics of the power distribution network.

[0027] In practical applications, this power metering device map model can be used for various scenarios such as power metering anomaly detection, load analysis, and line loss analysis. For example, in anomaly detection, by analyzing whether the node attributes (measured data) satisfy the relationships represented by the physical constraint edges, potential power metering anomalies can be discovered. Assuming the current measurements of M001, M002, M003, and M004 are 50A, 20A, 15A, and 10A respectively, according to Kirchhoff's current law, the current of M001 should be equal to the sum of the currents of M002, M003, and M004 (45A). However, the actual measured value is 50A, indicating a possible power metering anomaly or power theft.

[0028] In one optional implementation, in the power metering device graph model, multi-level association features between the target power metering device node and its adjacent nodes in the time and spatial dimensions are extracted through a graph structure propagation mechanism, resulting in association feature representations including: Centered on the target energy metering device node, extend outwards along the topological connection edge and the physical constraint edge by a preset propagation order to determine adjacent nodes that have a direct or indirect connection relationship with the target energy metering device node, thus obtaining a set of adjacent nodes; The change pattern of the measurement data of the target energy metering device node in the time series is obtained, the change pattern of the measurement data of each neighboring node in the neighboring node set in the time series is obtained, and the time synchronization and temporal correlation between the change pattern of the measurement data of the target energy metering device node and the change pattern of the measurement data of each neighboring node are calculated to obtain the time dimension correlation features; Based on the topological connection edges and physical constraint edges between the target energy metering device node and the set of adjacent nodes in the energy metering device graph model, the node attributes of each adjacent node in the set of adjacent nodes are propagated to the target energy metering device node along the edge relationships through a graph structure propagation mechanism. The propagated node attributes are weighted and aggregated according to the propagation path length and edge relationship type to obtain spatial dimension association features. The temporal dimension association features and the spatial dimension association features are fused to obtain the association feature representation.

[0029] To implement a method for extracting multi-level association features from a graph model of electricity metering equipment using a graph structure propagation mechanism, this implementation involves a graph-based feature extraction technique for electricity metering equipment. The method first constructs a graph model of electricity metering equipment, including electricity metering equipment nodes, topological connection edges, and physical constraint edges, and then extracts association features in the time and spatial dimensions using a graph structure propagation mechanism.

[0030] In practical applications, the power metering equipment network model consists of multiple power metering equipment nodes, which are interconnected through topological connection edges and physical constraint edges. Topological connection edges represent the network connection relationship between power metering equipment, such as the feeder connection between a transformer and a meter; physical constraint edges represent the physical constraint relationship between equipment, such as the geographical location constraint of equipment within the same substation. Each power metering equipment node contains static attributes such as equipment identifier, equipment type, and installation location, as well as dynamic attributes such as measured data like voltage, current, and active power.

[0031] When extracting associated features centered on the target electricity metering equipment node, the process first expands outward along the topological connection edges and physical constraint edges of the graph structure to identify adjacent nodes connected to the target node. For example, setting the propagation order to 2, for the target node A (such as a 10kV meter in a specific substation), the directly connected transformer node B (first-order adjacency) can be found through the topological connection edges, and then other meter nodes C and D (second-order adjacency) are connected through transformer B; simultaneously, other equipment nodes E and F within the same substation are found through the physical constraint edges. This yields the set of adjacent nodes {B, C, D, E, F}.

[0032] When acquiring the temporal correlation features between the target node and its neighboring nodes, the measurement data of each node over a period of time (e.g., the past 30 days) is first collected to form a time series. For meter A, its active power readings are collected every 15 minutes to obtain the time series {P_A1, P_A2, ..., P_An}; similarly, the corresponding time series of neighboring nodes B, C, D, E, and F are obtained. The changing patterns of the measurement data are extracted through time series analysis, including the shape of the daily load curve, the timing of peak and valley occurrences, and the amplitude of load fluctuations. For example, target meter A may exhibit a typical commercial electricity consumption pattern with peak daytime consumption from 9:00-11:30 and 14:00-17:00, and low nighttime consumption from 23:00-6:00.

[0033] When calculating time synchronization, the measurement data sequences of the target node A and its neighboring nodes are aligned within time windows to detect the time consistency of measurement value changes. For example, when the voltage of node A suddenly drops, does the directly connected node B also show a synchronous change? For time series correlation analysis, the correlation coefficient between the measurement data sequences of the target node and its neighboring nodes is calculated to identify strong correlations. For example, in an industrial area, meter A and meter C, which are powered by the same transformer, may show a correlation coefficient as high as 0.85, while meter F, which is on a different power supply circuit, has a correlation coefficient of only 0.32, indicating that the electricity consumption behaviors of A and C are highly correlated.

[0034] Spatial dimension association feature extraction is based on a graph structure propagation mechanism, which propagates the attribute information of adjacent nodes to the target node along the edges. During the propagation process, weights are applied according to the propagation path length and edge relationship type. For example, the influence weight of directly connected nodes (first order) can be set to 0.8, and the weight of indirectly connected nodes (second order) can be set to 0.5. Simultaneously, the propagation weight of topological connection edges may be higher than that of physical constraint edges; for example, the weight of a topological edge is 1.0, while the weight of a physical constraint edge is 0.6. For meter A, node attributes (such as transformer capacity and load rate) from the directly powered transformer B are propagated to A with a weight of 0.8 × 1.0 = 0.8; attributes from device E, which is in the same substation but not directly powered, are propagated to A with a weight of 0.8 × 0.6 = 0.48.

[0035] The node attributes propagated include static attributes such as device type, installation environment, and service life, as well as dynamic attributes such as load characteristics and voltage stability. For example, when target meter A is connected to multiple industrial meters, the characteristic of "industrial electricity" will be amplified through graph propagation; when target meter A is connected to multiple devices in areas with large voltage fluctuations, the characteristic of "unstable power grid area" will also be propagated and amplified.

[0036] Finally, the temporal and spatial correlation features are fused, which can be achieved through feature concatenation. For example, for target meter A, the temporal features might include "weekday double-peak electricity consumption pattern" and "high time-series correlation with node C (0.85)"; the spatial features might include "industrial area power supply" and "power supply from a large-capacity transformer (400kVA)". The fused correlation features are represented as a comprehensive feature vector of A, containing correlation features in both temporal and spatial dimensions, providing rich feature representations for subsequent tasks such as anomaly detection of electricity metering equipment and load forecasting.

[0037] The multidimensional correlation features extracted by the above methods can comprehensively capture the spatiotemporal correlation of power metering equipment in the power network, effectively supporting the intelligent operation and maintenance and management of the power system.

[0038] In one optional implementation, the anomaly detection model calculates the degree of deviation of the measurement data under the constraints of the associated feature representation. When the degree of deviation exceeds a preset deviation threshold, it is determined that the measurement data is abnormal, and the anomaly detection result includes: The associated feature representation is input into the anomaly detection model, and the anomaly detection model performs prediction calculations on the measurement data of the target power metering equipment node based on the associated feature representation to obtain the predicted measurement data; Calculate the difference metric between the actual measurement data and the predicted measurement data of the target energy metering device node. The difference metric characterizes the degree to which the actual measurement data deviates from the normal value range constrained by the correlation feature representation, and obtain the degree of deviation. The deviation degree is compared with the preset deviation threshold. When the deviation degree exceeds the preset deviation threshold, it is determined that the measurement data of the target power metering device node is abnormal and an abnormality mark is generated. When the deviation degree does not exceed the preset deviation threshold, it is determined that the measurement data of the target power metering device node is normal and a normality mark is generated, thus obtaining the abnormality detection result.

[0039] In this embodiment, historical measurement data from multiple energy metering device nodes in the power grid are first acquired. A target energy metering device node is selected, and an anomaly detection model is trained based on the correlation between the multiple energy metering device nodes. For the measurement data of the target energy metering device node, anomaly detection is performed by constructing a correlation feature representation and calculating the degree of deviation using the anomaly detection model.

[0040] Specifically, the step of calculating the deviation of measurement data through the constraint of correlation feature representation includes inputting the correlation feature representation into the anomaly detection model, and based on this, performing predictive calculations on the measurement data of the target electricity metering device node to obtain the predicted measurement data. For example, if the correlation feature representation contains historical electricity consumption data of four electricity metering devices A, B, C, and D, which are 125kWh, 98kWh, 156kWh, and 205kWh respectively, and the target node E has a strong correlation with these four nodes, then the anomaly detection model will predict that the normal electricity consumption of node E should be 178kWh based on the data of these four nodes.

[0041] Subsequently, a difference metric is calculated between the actual and predicted measurement data of the target energy metering device node. This difference metric characterizes the degree to which the actual measurement data deviates from the normal range constrained by the associated characteristic representation, thus yielding the degree of deviation. In practical applications, the difference metric can be calculated using methods such as absolute difference, relative difference, or statistical distribution distance. For example, if the actual measurement data of target node E is 240 kWh and the predicted measurement data is 178 kWh, then the absolute difference is |240-178|=62 kWh, and the relative difference is |240-178| / 178≈0.348 or 34.8%.

[0042] The calculated deviation is compared with a preset deviation threshold. If the deviation exceeds the preset threshold, the system determines that the measurement data of the target energy metering device node is abnormal and generates an anomaly flag. If the deviation does not exceed the preset threshold, the system determines that the measurement data of the target energy metering device node is normal and generates a normal flag, thus obtaining the anomaly detection result. For example, if the preset relative difference threshold is 15%, and the calculated relative difference is 34.8%, which significantly exceeds the threshold, the system will determine that the measurement data of node E is abnormal and generate a corresponding anomaly flag.

[0043] In practical deployments, the preset deviation threshold can be set based on historical data analysis and expert experience, or it can be adaptively determined through statistical methods. For example, for a residential meter group in a power distribution network, if analysis of three months of historical data reveals that the normal fluctuation range is usually within ±10%, then the preset deviation threshold can be set to 10%. For industrial electrical equipment, due to different load characteristics, a deviation threshold of 20% may be necessary.

[0044] Furthermore, to improve detection accuracy, the time factor can be considered when calculating the difference metric. For example, for a commercial area's electricity meter, there are significant differences in electricity consumption patterns between weekdays and weekends; direct comparison may lead to false alarms. By introducing time-dimensional correlation features, the system can automatically distinguish between weekday and weekend data for prediction. For instance, if the typical electricity consumption of the associated node on weekdays is 150kWh, 120kWh, 175kWh, and 220kWh, the predicted target node should be 185kWh; while on weekends, the electricity consumption of the associated node drops to 90kWh, 75kWh, 110kWh, and 140kWh, the predicted target node should be 120kWh.

[0045] This solution can also differentiate between anomalies of varying severity by setting multi-level deviation thresholds. For example, a threshold of 15% can be set for minor anomalies, 30% for moderate anomalies, and 50% for severe anomalies. When a meter's deviation is 25%, the system classifies it as a minor anomaly; a deviation of 40% is classified as a moderate anomaly, which may require manual verification; and a deviation exceeding 50% is classified as a severe anomaly, at which point the system will immediately trigger an alarm mechanism and notify relevant personnel for handling.

[0046] In anomaly detection, the quality of the constructed association feature representation directly affects the detection effect. In practice, it has been found that selecting 5-8 strongly correlated neighboring nodes to construct association features usually achieves good results. For example, in a residential community's electricity meter group, data from 6 meters located in the same building were selected to construct association feature representations. These meter users have similar daily routines and electricity usage habits, allowing the system to effectively detect abnormal electricity consumption.

[0047] Furthermore, anomaly detection models can be implemented using different machine learning algorithms, such as support vector machines, random forests, and deep neural networks, depending on the specific application requirements. In one example, a three-layer neural network structure was employed: the input layer receives data from multiple nodes in the associated feature representation, the hidden layer performs feature transformation and extraction, and the output layer generates predicted measurement data. Through training and validation on historical data, the model achieved an anomaly detection accuracy of 93.5% on the test set.

[0048] The above technical solutions enable efficient anomaly detection of power metering equipment node measurement data in the power grid, effectively identify anomalies in power data, and provide strong support for applications such as power grid monitoring, power theft identification, and equipment fault early warning.

[0049] In one optional implementation, the anomaly detection model performs predictive calculations on the measurement data of the target energy metering device node based on the associated feature representation to obtain the predicted measurement data, which includes: Based on the historical measurement data of the target energy metering device node and the time dimension correlation features in the correlation feature representation, the initial prediction measurement data of the target energy metering device node is generated through the initial prediction module of the anomaly detection model; The difference between the initial predicted measurement data and the historical measurement data of the target power metering equipment node is calculated, and the residual information that the initial predicted measurement data fails to represent is extracted to obtain the prediction residual characteristics; The predicted residual features and the spatial dimension correlation features in the correlation feature representation are input into the graph convolution correction module of the anomaly detection model. The predicted residual features are then propagated along the topological connection edges and physical constraint edges in the power metering equipment graph model through the graph convolution correction module. The predicted residual features are corrected based on the spatial constraint relationship of adjacent nodes to obtain the corrected residual value. The initial predicted measurement data is superimposed with the corrected residual value to obtain the predicted measurement data.

[0050] The initial prediction module of the anomaly detection model acquires historical measurement data of the target energy metering equipment node. For example, it acquires hourly current data of the A-phase current sensor of a substation over the past 30 days, totaling 720 historical data points. Simultaneously, it extracts time-related features associated with the sensor from the associated feature representation, such as weekday / rest day markings, 24-hour cycle features, and temperature change features. For instance, the period from 9:00 to 18:00 from Monday to Friday is marked as the weekday peak period, while the entire weekend and other weekday periods are marked as the off-peak period. Temperature change data over the past 30 days is used as the temperature feature. The initial prediction module processes this input data using a time-series prediction algorithm to generate initial predicted measurement data for the target energy metering equipment node. Specifically, based on the changing patterns of historical data for the same period and the influence of time features, this module can predict the expected reading of the A-phase current sensor for the next hour, for example, predicting a current value of 95.3 amperes for the next hour.

[0051] Next, the system calculates the difference between the initial predicted measurement data and the actual historical measurement data of the target power metering equipment node, extracting residual information that the initial predicted measurement data failed to fully represent. For example, assuming the actual historical current value for the previous hour was 100.5 amps, while the initial predicted value was 95.3 amps, the calculated residual would be 5.2 amps. By analyzing the patterns and distribution characteristics of recent predicted residuals, the system extracts predicted residual features. These features may include the magnitude, trend, and periodic fluctuations of the residuals. For instance, if the sensor's residuals have consistently been positive and fluctuated between 4 and 6 amps over the past 5 days, it indicates that the initial prediction module may have systematically underestimated the value.

[0052] Subsequently, the system inputs the predicted residual features along with the spatial dimension correlation features in the correlation feature representation into the graph convolution correction module of the anomaly detection model. The spatial dimension correlation features include the physical connection relationships between power metering devices and power balance constraints. For example, the A-phase current sensor is topologically adjacent to the B-phase and C-phase current sensors, bus voltage sensors, and other devices within the same substation, and the three-phase currents should maintain relative balance under normal conditions. The graph convolution correction module propagates the predicted residual features along the topological connection edges and physical constraint edges in the power metering device graph model. This propagation process considers the grid topology and the physical constraints of the power system. For example, if adjacent B-phase and C-phase current sensors also show similar underestimation of amplitude, it indicates that this may be a systematic deviation caused by common factors, and the correction module will adjust according to the three-phase balance principle; if other phases do not show similar deviations, it may be a situation unique to phase A. Through message passing and aggregation on the graph structure, the system corrects the predicted residual features based on the spatial constraint relationships of adjacent nodes to obtain the corrected residual value. In this example, the corrected residual value may be adjusted from the original 5.2 amperes to 4.8 amperes because, taking into account the three-phase balance, the system assumes that some of the residuals may come from temporary fluctuations rather than systematic deviations.

[0053] Finally, the system superimposes the initial predicted measurement data with the corrected residual value to obtain the final predicted measurement data. For example, adding the initial predicted value of 95.3 amperes to the corrected residual value of 4.8 amperes yields the final predicted measurement data of 100.1 amperes, which is closer to the actual historical value of 100.5 amperes, demonstrating the effectiveness of spatial constraint correction.

[0054] In practical applications, for example, a power distribution network contains 100 power metering device nodes, forming a complex graph structure. The system collects historical measurement data for each node over the past 90 days, sampling one point every 15 minutes, for a total of 8640 time points. For the target node (such as the power metering point of a 10kV distribution line), the initial prediction module generates an initial prediction value of 5.23MW based on its historical load curve and time characteristics (such as weekday markings, temperature data, etc.). The difference from the actual value of 5.48MW is calculated, yielding a prediction residual of 0.25MW. The graph convolution correction module analyzes the topological relationship and power balance constraints between the node and its connected transformers, buses, and other distribution lines, finding an unallocated difference of approximately 0.3MW between the total output power of the upstream transformer and the sum of the power of each distribution line, and this difference should primarily be attributed to the target node. Through two rounds of graph convolution, the system corrects the prediction residual to 0.27MW, ultimately obtaining a corrected predicted measurement of 5.50MW. The deviation from the actual value is reduced to 0.02MW, significantly improving prediction accuracy. This method fully utilizes the temporal and spatial correlations within the power system, effectively enhancing the accuracy of electricity metering data prediction.

[0055] In one optional implementation, the data quality assessment index of the target power metering device node is calculated based on the frequency of anomalies and the duration of anomalies, including: The frequency of anomalies occurring at the target energy metering equipment nodes within a preset evaluation time window is statistically analyzed, and the ratio of the anomaly frequency to the total number of samples within the preset evaluation time window is calculated to obtain the anomaly frequency percentage. The abnormal duration of all abnormal events at the target energy metering device node within the preset evaluation time window is statistically analyzed, and the ratio of the abnormal duration to the total duration of the preset evaluation time window is calculated to obtain the abnormal duration percentage. The frequency impact weight is determined based on the degree of influence of the abnormality frequency ratio on data quality, and the duration impact weight is determined based on the degree of influence of the abnormality duration ratio on data quality. The sum of the frequency impact weight and the duration impact weight is a normalized value. The frequency quality loss value is obtained by weighting the abnormal frequency ratio with the frequency impact weight, and the duration quality loss value is obtained by weighting the abnormal duration ratio with the duration impact weight. The frequency quality loss value and the duration quality loss value are aggregated to obtain the total quality loss value. The data quality assessment index of the target power metering equipment node is calculated based on the total quality loss value.

[0056] This embodiment relates to a data quality assessment method for electricity metering equipment nodes. In practical applications, electricity metering equipment nodes may experience data anomalies, which can affect the accuracy of power grid monitoring and electricity data analysis. To accurately assess the data quality of electricity metering equipment nodes, this embodiment proposes a data quality assessment method based on the frequency and duration of anomalies.

[0057] In this implementation, the first step is to acquire historical data from the target electricity metering device node. This data includes various measurement data generated during the operation of the electricity metering device, such as voltage, current, and power. By analyzing this historical data, events indicating data anomalies can be identified, including missing data, abnormal data values, and abnormal data fluctuations.

[0058] For identified abnormal events, this implementation method evaluates them from two dimensions: the frequency of abnormal occurrence and the duration of abnormality. The frequency of abnormal occurrences at the target energy metering equipment node within a preset evaluation time window is statistically analyzed. For example, if the evaluation time window is set to 30 days, and the equipment performs 4320 data samples during these 30 days (assuming sampling every 10 minutes), with 216 samples exhibiting abnormalities, then the abnormal occurrence frequency is 216. The ratio of the abnormal occurrence frequency to the total number of samples within the preset evaluation time window is calculated to obtain the abnormal frequency percentage. In the above example, the abnormal frequency percentage is 216 / 4320 = 0.05, or 5%.

[0059] Simultaneously, the duration of all abnormal events at the target electricity metering equipment nodes within the preset evaluation time window is statistically analyzed. The duration of an abnormal event refers to the time interval from the start to the end of the abnormal event. For example, within a 30-day evaluation time window, the cumulative duration of all abnormal events is 1800 minutes. The ratio of the duration of abnormal events to the total duration of the preset evaluation time window is calculated to obtain the percentage of abnormal duration. In the example above, the total duration of the preset evaluation time window is 30 days × 24 hours × 60 minutes = 43200 minutes, and the percentage of abnormal duration is 1800 / 43200 = 0.0417, or 4.17%.

[0060] The frequency impact weight is determined based on the degree of influence of the anomaly frequency percentage on data quality. In practical applications, the impact of anomaly frequency on data quality may vary depending on the scenario. For example, in scenarios requiring high-precision real-time monitoring, frequent anomalies may lead to monitoring failure, thus the frequency impact weight is high; while in scenarios focusing only on long-term trends, short-term anomalies have a smaller impact, and the frequency impact weight is relatively low. Assuming that in the current application scenario, the frequency impact weight is determined to be 0.6.

[0061] The weight of duration is determined based on the degree of impact of the proportion of anomalies on data quality. The impact of anomaly duration on data quality also varies depending on the scenario. For example, in scenarios requiring continuous data analysis, prolonged data anomalies can severely affect the analysis results, thus the weight of duration is high; while in scenarios focusing only on data at a specific point in time, the impact of anomaly duration is relatively small, and the weight of duration is low. Assuming that in the current application scenario, the weight of duration is determined to be 0.4.

[0062] It is important to note that the sum of the frequency impact weight and the duration impact weight is a normalized value, which equals 1. This ensures the standardization of the evaluation indicators and facilitates comparisons between different device nodes.

[0063] The frequency quality loss value is calculated by weighting the percentage of abnormal frequencies with the frequency impact weight. In the example above, the frequency quality loss value is 0.05 × 0.6 = 0.03. The duration quality loss value is calculated by weighting the percentage of abnormal durations with the duration impact weight. In the example above, the duration quality loss value is 0.0417 × 0.4 = 0.01668.

[0064] The frequency quality loss value and the duration quality loss value are aggregated to obtain the total quality loss value. The aggregation method can be simple summation or other more complex aggregation functions. In the case of simple summation, the total quality loss value is 0.03 + 0.01668 = 0.04668.

[0065] The data quality assessment index for the target energy metering equipment node is calculated based on the total quality loss. The data quality assessment index can be defined as 1 minus the total quality loss, representing the validity or reliability of the data. In the example above, the data quality assessment index is 1 - 0.04668 = 0.95332, which is approximately 95.33%.

[0066] This assessment method considers both the frequency and duration of anomalies, providing a more comprehensive reflection of the data quality status of electricity metering equipment nodes. Data quality assessment indicators can be used to monitor the operational status of electricity metering equipment, identify equipment requiring maintenance or replacement, and improve the efficiency and reliability of the power grid. Furthermore, this method can also be applied to the performance evaluation and optimization of electricity metering systems, providing decision support for power grid operations.

[0067] In practical applications, the values ​​of frequency and duration influence weights can be adjusted according to different application scenarios and needs to obtain data quality assessment results that better meet the requirements of specific scenarios. Furthermore, the assessment dimensions can be expanded, such as considering the severity of anomalies and their impact on downstream applications, to further improve the accuracy and practicality of the assessment.

[0068] The smart meter data anomaly detection and quality assessment optimization system of this invention includes: The first unit is used to acquire measurement data of multiple energy metering devices over time, perform node mapping on the multiple energy metering devices based on the topological connection relationship and physical constraint relationship of the power distribution network, use the measurement data as node attributes, and use the topological connection relationship and physical constraint relationship as edge relationship to construct an energy metering device graph model; The second unit is used to extract multi-level association features between the target energy metering device node and its adjacent nodes in the time and space dimensions through a graph structure propagation mechanism in the energy metering device graph model, and obtain the association feature representation. The third unit is used to input the correlation feature representation and the measurement data of the target power metering device node into the anomaly detection model, calculate the degree of deviation of the measurement data under the constraint of the correlation feature representation through the anomaly detection model, and determine that the measurement data is abnormal when the degree of deviation exceeds a preset deviation threshold, thereby obtaining the anomaly detection result; The fourth unit is used to count the frequency and duration of anomalies of the target power metering device node within a preset time window based on the anomaly detection results, and to calculate the data quality evaluation index of the target power metering device node based on the frequency and duration of anomalies.

[0069] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0070] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0071] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent electric meter data anomaly detection and quality evaluation optimization method, characterized in that, The method comprises the following steps: obtaining measurement data of a plurality of electric energy metering devices in a time sequence, mapping the plurality of electric energy metering devices based on a topological connection relationship and a physical constraint relationship of a power distribution network, taking the measurement data as node attributes, taking the topological connection relationship and the physical constraint relationship as edge relationships, and constructing an electric energy metering device graph model; in the electric energy metering device graph model, a multi-order correlation feature between a target electric energy metering device node and its adjacent nodes in the time dimension and the space dimension is extracted through a graph structure propagation mechanism to obtain a correlation feature representation; the correlation feature representation and the measurement data of the target electric energy metering device node are input into an anomaly detection model, the deviation degree of the measurement data under the constraint of the correlation feature representation is calculated through the anomaly detection model, and when the deviation degree exceeds a preset deviation threshold, it is determined that the measurement data is abnormal, and an anomaly detection result is obtained; according to the anomaly detection result, the abnormal occurrence frequency and the abnormal duration of the target electric energy metering device node in a preset time window are counted, and the data quality evaluation index of the target electric energy metering device node is calculated based on the abnormal occurrence frequency and the abnormal duration.

2. The method of claim 1, wherein, mapping the plurality of electric energy metering devices based on the topological connection relationship and the physical constraint relationship of the power distribution network, taking the measurement data as node attributes, taking the topological connection relationship and the physical constraint relationship as edge relationships, and constructing an electric energy metering device graph model comprises: analyzing the topological connection relationship of the power distribution network, determining the upstream and downstream connection structure and branch attribution relationship between the plurality of electric energy metering devices, and obtaining a topological connection structure; extracting the physical constraint relationship between the plurality of electric energy metering devices based on the topological connection structure, and obtaining a physical constraint structure; mapping the plurality of electric energy metering devices as graph nodes, taking the voltage measurement data, current measurement data and power measurement data in the measurement data as node attributes of the corresponding graph nodes, and constructing a node set containing the graph nodes corresponding to the plurality of electric energy metering devices; establishing a topological connection edge between the graph nodes with upstream and downstream connection relationship according to the topological connection structure, and establishing a physical constraint edge between the graph nodes with physical law constraint according to the physical constraint structure, to obtain an edge set; generating the electric energy metering device graph model based on the node set and the edge set, and the topological connection edge and the physical constraint edge in the electric energy metering device graph model jointly represent the correlation relationship between the plurality of electric energy metering devices.

3. The method of claim 1, wherein, in the electric energy metering device graph model, a multi-order correlation feature between a target electric energy metering device node and its adjacent nodes in the time dimension and the space dimension is extracted through a graph structure propagation mechanism to obtain a correlation feature representation, which comprises: centering on the target electric energy metering device node, expanding outward along the topological connection edge and the physical constraint edge by a preset propagation order, determining adjacent nodes having a direct connection relationship or an indirect connection relationship with the target electric energy metering device node, and obtaining an adjacent node set; obtain a change pattern of the measurement data of the target electric energy metering device node in a time sequence, obtain a change pattern of the measurement data of each adjacent node in the set of adjacent nodes in a time sequence, calculate a time synchronization and a time sequence correlation between the change pattern of the measurement data of the target electric energy metering device node and the change pattern of the measurement data of each adjacent node, and obtain a time-dimension correlation feature; based on the topological connection edges and the physical constraint edges between the target electric energy metering device node and the set of adjacent nodes in the electric energy metering device graph model, propagate the node attributes of each adjacent node in the set of adjacent nodes to the target electric energy metering device node through a graph structure propagation mechanism, and aggregate the propagated node attributes according to a propagation path length and an edge relationship type to obtain a space-dimension correlation feature; perform feature fusion on the time-dimension correlation feature and the space-dimension correlation feature to obtain the correlation feature representation.

4. The method of claim 1, wherein, calculate a deviation degree of the measurement data under the constraint of the correlation feature representation through the anomaly detection model, determine that the measurement data is abnormal when the deviation degree exceeds a preset deviation threshold, and obtain an anomaly detection result including: input the correlation feature representation into the anomaly detection model, and perform prediction calculation on the measurement data of the target electric energy metering device node based on the correlation feature representation through the anomaly detection model to obtain predicted measurement data; calculate a difference measure value between the actual measurement data of the target electric energy metering device node and the predicted measurement data, the difference measure value representing a degree to which the actual measurement data deviates from a normal value range constrained by the correlation feature representation, and obtain the deviation degree; compare the deviation degree with the preset deviation threshold, determine that the measurement data of the target electric energy metering device node is abnormal and generate an abnormal identifier when the deviation degree exceeds the preset deviation threshold, and determine that the measurement data of the target electric energy metering device node is normal and generate a normal identifier when the deviation degree does not exceed the preset deviation threshold, and obtain the anomaly detection result.

5. The method of claim 4, wherein, perform prediction calculation on the measurement data of the target electric energy metering device node based on the correlation feature representation through the anomaly detection model to obtain predicted measurement data including: based on the historical measurement data of the target electric energy metering device node and the time-dimension correlation feature in the correlation feature representation, generate initial predicted measurement data of the target electric energy metering device node through an initial prediction module of the anomaly detection model; perform difference calculation on the initial predicted measurement data and the historical measurement data of the target electric energy metering device node, extract residual information that the initial predicted measurement data fails to represent, and obtain a prediction residual feature; input the predicted residual feature into a graph convolution correction module of the anomaly detection model, propagate the predicted residual feature along the topological connection edges and the physical constraint edges in the electric energy metering device graph model through the graph convolution correction module, correct the predicted residual feature based on the spatial constraint relationship of adjacent nodes to obtain a corrected residual value; superimpose the initial predicted measurement data and the corrected residual value to obtain the predicted measurement data.

6. The method of claim 1, wherein, calculating the data quality evaluation index of the target electric energy metering device node based on the abnormal occurrence frequency and the abnormal duration includes: statistically counting the abnormal occurrence frequency of the target electric energy metering device node within a preset evaluation time window, calculating the ratio of the abnormal occurrence frequency to the total sampling number within the preset evaluation time window to obtain an abnormal frequency proportion; statistically counting the abnormal duration of all abnormal events of the target electric energy metering device node within the preset evaluation time window, calculating the ratio of the abnormal duration to the total duration of the preset evaluation time window to obtain an abnormal duration proportion; determining a frequency influence weight based on the influence degree of the abnormal frequency proportion on data quality, determining a duration influence weight based on the influence degree of the abnormal duration proportion on data quality, and the sum of the frequency influence weight and the duration influence weight is a normalization value; performing weighted calculation on the abnormal frequency proportion and the frequency influence weight to obtain a frequency quality loss value, performing weighted calculation on the abnormal duration proportion and the duration influence weight to obtain a duration quality loss value, aggregating the frequency quality loss value and the duration quality loss value to obtain a total quality loss value, and calculating the data quality evaluation index of the target electric energy metering device node based on the total quality loss value.

7. A system for smart meter data anomaly detection and quality assessment optimization, configured to implement the method of any one of claims 1-6, characterized in that, comprises: a first unit configured to acquire measurement data of a plurality of electric energy metering devices in a time sequence, map the plurality of electric energy metering devices based on topological connection relationships and physical constraint relationships of a power distribution network, take the measurement data as node attributes, take the topological connection relationships and physical constraint relationships as edge relationships, and construct an electric energy metering device graph model; a second unit configured to extract, in the electric energy metering device graph model, multi-order correlation features between a target electric energy metering device node and its adjacent nodes in a time dimension and a spatial dimension through a graph structure propagation mechanism to obtain a correlation feature representation; a third unit configured to input the correlation feature representation and measurement data of the target electric energy metering device node into an anomaly detection model, calculate a deviation degree of the measurement data under the constraint of the correlation feature representation through the anomaly detection model, determine that the measurement data is abnormal when the deviation degree exceeds a preset deviation threshold, and obtain an anomaly detection result; a fourth unit configured to statistically count an abnormal occurrence frequency and an abnormal duration of the target electric energy metering device node within a preset time window according to the anomaly detection result, and calculate a data quality evaluation index of the target electric energy metering device node based on the abnormal occurrence frequency and the abnormal duration.

8. An electronic device, comprising: comprises: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the instructions stored by the memory to perform the method of any one of claims 1 to 6.

9. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions, when executed by a processor, implement the method of any one of claims 1 to 6.

Images