Electric energy meter implicit fault prediction method and system based on time sequence feature clustering

Abstract

This invention provides a method and system for predicting latent faults in electricity meters based on time-series feature clustering, belonging to the field of data analysis technology. The method includes: acquiring multiple operational datasets, performing multi-scale time-series feature extraction and clustering, and establishing multiple electricity meter state spaces; performing anomaly gradient detection to obtain multiple electricity meter anomaly gradient matrices; performing multi-dimensional latent fault prediction based on a latent fault mechanism library to obtain the fault prediction domain for each electricity meter; performing state transition analysis based on time decay weights to establish multiple latent fault transition relationship models; and compensating for group evolution effects based on a group relationship set to construct a group state evolution model for electricity meters. This invention solves the technical problem of existing technologies relying on static fault detection mechanisms, which typically rely on simple thresholds or preset standards of electricity meters to determine whether a fault has occurred, failing to accurately identify complex and latent faults, thus affecting the stability of the power system.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to a method and system for predicting latent faults in electricity meters based on time-series feature clustering. Background Technology

[0002] Electricity meters are an indispensable core device in modern power systems. Their main function is to accurately measure and record electricity consumption data. With the increasing intelligence and complexity of power systems, electricity meters may develop hidden faults during long-term operation. These hidden faults are often not easily detected by traditional detection methods, but they may have a serious impact on the normal operation of the power system and the accuracy of data. Especially in large-scale power systems, they may lead to metering errors, statistical deviations, or even the instability of the entire system.

[0003] Existing technologies rely on static fault detection mechanisms, which typically rely on simple thresholds or preset standards of electricity meters to determine whether a fault has occurred. This static threshold-based detection method cannot accurately identify complex and hidden faults, especially when equipment is gradually aging or experiencing minor faults. Faults such as meter drift and component aging are often gradual and invisible, and traditional threshold methods have difficulty capturing these subtle changes. This may lead to instability in the operation of the power system and affect the quality and efficiency of power supply. Summary of the Invention

[0004] This application provides a method and system for predicting latent faults in electricity meters based on time-series feature clustering. It aims to solve the technical problem that existing technologies rely on static fault detection mechanisms, which usually rely on simple thresholds or preset standards of electricity meters to determine whether a fault has occurred. This fails to accurately identify complex and latent faults, thereby affecting the stability of the power system.

[0005] The first aspect disclosed in this application provides a method for predicting latent faults in electricity meters based on time-series feature clustering. The method includes: acquiring multiple operational datasets of multiple electricity meters within a target area, and performing multi-scale time-series feature extraction and time-series feature clustering on the multiple operational datasets to establish multiple electricity meter state spaces; performing anomaly gradient detection on the multiple electricity meter state spaces to obtain multiple electricity meter anomaly gradient matrices; performing multi-dimensional latent fault prediction on each electricity meter anomaly gradient matrix based on a latent fault mechanism library to obtain fault prediction domains for each electricity meter; performing state transition analysis based on time decay weights on each electricity meter fault prediction domain to establish multiple latent fault transition relationship models; and compensating for group evolution effects on the multiple latent fault transition relationship models based on a group relationship set among the multiple electricity meters to construct an electricity meter group state evolution model.

[0006] The second aspect of this application discloses a system for predicting latent faults in electricity meters based on time-series feature clustering. The system is used in the aforementioned method for predicting latent faults in electricity meters based on time-series feature clustering. The system includes: a time-series feature clustering module, used to acquire multiple operational datasets of multiple electricity meters within a target area, and to perform multi-scale time-series feature extraction and time-series feature clustering on the multiple operational datasets to establish multiple electricity meter state spaces; an anomaly gradient detection module, used to perform anomaly gradient detection on the multiple electricity meter state spaces to obtain multiple electricity meter anomaly gradient matrices; a latent fault prediction module, used to perform multi-dimensional latent fault prediction on each electricity meter anomaly gradient matrix based on a latent fault mechanism library to obtain fault prediction domains for each electricity meter; a state transition analysis module, used to perform state transition analysis based on time decay weights on the fault prediction domains of each electricity meter to establish multiple latent fault transition relationship models; and a group evolution effect compensation module, used to compensate for group evolution effects on the multiple latent fault transition relationship models based on a group relationship set among the multiple electricity meters to construct a group state evolution model for electricity meters.

[0007] One or more technical solutions provided in this application have at least the following beneficial effects:

[0008] By extracting multi-scale time-series features from the operational data of multiple electricity meters, the operating characteristics of electricity meters at different time scales can be comprehensively captured. Multi-scale features enable a deeper understanding of the operating status of electricity meters, thereby more accurately predicting potential latent faults. Through anomaly gradient detection, the state changes of electricity meters can be monitored in real time, identifying subtle but potentially faulty gradient trends and generating anomaly gradient matrices. This allows for early detection of potential faults and the implementation of preventative measures. By utilizing a latent fault mechanism library and combining it with the electricity meter anomaly gradient matrix, multi-dimensional fault prediction can be performed. Targeted predictions can be made based on different fault types of electricity meters, providing inferences of multiple potential fault paths, thus enhancing the accuracy of fault prediction. By correcting for time decay weights, a latent fault transfer relationship model can be established by incorporating the time effect during fault transfer in state transition analysis. This model not only focuses on the occurrence of faults but also on the fault propagation process, ensuring the dynamic nature of fault prediction. By compensating for the group evolution effect of the fault transfer relationship model of the electricity meter through the group relationship set, a group state evolution model reflecting the behavior of the entire electricity meter group is finally constructed. This model incorporates the mutual influence between electricity meters and can accurately identify potential fault propagation paths between devices in the group. Through the group state evolution model, the impact of certain electricity meter faults on other devices can be predicted, thereby enabling early response and reducing fault chain reactions and system downtime.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the method for predicting latent faults in electricity meters based on time-series feature clustering, provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the structure of a latent fault prediction system for electricity meters based on time-series feature clustering, provided in an embodiment of this application.

[0012] Figure labeling: Temporal feature clustering module 10, anomaly gradual change detection module 20, latent fault prediction module 30, state transition analysis module 40, population evolution effect compensation module 50. Detailed Implementation

[0013] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0014] Example 1, as Figure 1 As shown in the embodiments of this application, a method for predicting latent faults in electricity meters based on time-series feature clustering is provided. The method includes: Multiple operational datasets of multiple electricity meters within the target area are obtained, and multi-scale time-series feature extraction and time-series feature clustering are performed on the multiple operational datasets to establish multiple electricity meter state spaces.

[0015] Operational datasets are obtained from multiple electricity meters within the target area. Specifically, operational data from the electricity meters is continuously or periodically collected via wireless sensors, communication networks, or physical interfaces. These datasets contain various operational indicators of the electricity meters over different time periods, such as voltage, current, and power. Temporal features are extracted from the operational datasets. To capture features at different time scales, multi-scale methods are used to extract short-, medium-, and long-term features. Cluster analysis is then performed on the extracted multi-scale temporal features using methods such as K-means clustering and DBSCAN to classify the operational status of the electricity meters into different categories, such as normal, slightly abnormal, and severely abnormal. Based on the clustering results, a state space for multiple electricity meters is established, mapping the operational status of the electricity meters to different regions in the space. This allows for accurate quantification and visualization of the operational status of the electricity meters.

[0016] Anomaly gradient detection is performed on the state space of the multiple energy meters to obtain multiple energy meter anomaly gradient matrices.

[0017] Anomaly gradient detection is performed on the state space of each electricity meter to identify the trend of state changes, including the direction of state change, the rate of state change, and the magnitude of state change. Through these three identifications, the abnormal gradient process of the electricity meter state is clarified, and these abnormal gradients are precursors to latent faults. The detected abnormal gradient information is organized into an anomaly gradient matrix to provide input for subsequent fault prediction and help determine the occurrence of latent faults.

[0018] Based on the latent fault mechanism library, multidimensional latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the fault prediction domain of each energy meter.

[0019] The latent fault mechanism library contains various mechanisms related to electricity meter faults, such as metering drift mechanisms, component aging mechanisms, and sampling distortion mechanisms. These mechanisms describe different causes and manifestations of electricity meter faults. Based on the anomaly gradient matrix of each electricity meter, the latent fault mechanism library is consulted and predictions are made. For example, if the metering drift mechanism of an electricity meter matches its state gradient, then the mechanism is matched with the anomaly gradient to predict that the electricity meter will experience a latent fault of metering drift. Each electricity meter fault prediction domain represents the possible fault types and fault paths that the electricity meter may experience in the future, providing a predictive basis for subsequent fault diagnosis and repair.

[0020] A state transition analysis based on time decay weights is performed on the fault prediction domain of each electricity meter to establish multiple hidden fault transfer relationship models.

[0021] Since the evolution of fault states changes over time, a time decay weight is used to adjust the transfer trend; that is, the closer to the current time, the greater the impact of the transfer, and vice versa. The process of transitioning from a normal state to a fault state in an energy meter is analyzed to predict the possible state changes each energy meter may experience within a certain future timeframe. By modeling the fault transfer trend of each energy meter, multiple implicit fault transfer relationship models are obtained, which indicate the dynamic process of fault occurrence.

[0022] Based on the group relationship set among the multiple energy meters, the group evolution effect compensation is performed on the multiple latent fault transfer relationship models to construct an energy meter group state evolution model.

[0023] The group relationship set includes the spatial topological relationships, historical fault relationships, and state change synchronization relationships among electricity meters. These relationships describe the mutual influence and dependence of multiple electricity meters during operation. Based on the group relationships among electricity meters, the implicit fault transfer relationship model is compensated, and the fault transfer models of different electricity meters are adjusted. Through this compensation, the group effect can be captured, that is, how the fault of one electricity meter affects the operating state of other electricity meters. Finally, through the compensation effect, a group state evolution model of electricity meters is constructed. This model can reflect the dynamic state changes of the entire group of electricity meters, including the propagation of faults, the scope of influence, and their evolution process.

[0024] Furthermore, multi-scale time-series feature extraction and time-series feature clustering are performed on the multiple operational datasets to establish multiple energy meter state spaces, including: Each running dataset is segmented according to a preset time window to obtain a set of running segments for each electricity meter; features are extracted from the set of running segments for each electricity meter based on multi-scale time-series features to obtain a feature set for each electricity meter; features are fused from the feature sets for each electricity meter to generate a time-series feature sequence for each electricity meter; time-series feature clustering is performed on the time-series feature sequences for each electricity meter to obtain multiple sets of running state categories; and association mapping is performed on the time-series feature sequences for each electricity meter based on the multiple sets of running state categories to generate a state space for the multiple electricity meters.

[0025] A predefined time window is defined to divide the continuous operating data of the electricity meter into multiple time periods, each representing an operating segment. These include: short-cycle operating segments, which refer to the fluctuation data of the electricity meter within a short period, used to capture short-term load fluctuations and rapid changes; medium-cycle operating segments, which involve the medium-cycle fluctuations of the electricity meter during normal operation, reflecting the medium- to long-term stability of the equipment; and long-cycle operating segments, which are related to the long-term trends of the electricity meter, such as equipment aging and long-term metering drift. By segmenting the data using time windows, a set of operating segments for each electricity meter is generated, with each meter corresponding to a series of operating segments at different time scales.

[0026] Feature extraction is performed on each electricity meter's operating segment, particularly at different scales of short, medium, and long cycles, extracting time-series features, including: time-domain features such as the fluctuation amplitude, mean, maximum, and minimum values ​​of current and voltage; frequency-domain features such as frequency components and harmonic distortion; statistical features such as standard deviation, skewness, and kurtosis; and trend features such as linear trends and drift trends. After feature extraction for each electricity meter's operating segment, a set of electricity meter features is generated, reflecting the electricity meter's behavior patterns at different time scales.

[0027] Feature sets extracted from different time scales, such as short-term, medium-term, and long-term, are fused. This can be achieved through various methods, such as: weighted averaging, which assigns different weights to features at different time scales and performs a weighted average; concatenation fusion, which concatenates feature vectors extracted from different periods to form a comprehensive feature vector; and principal component analysis, which fuses multiple feature sets through dimensionality reduction methods to extract the most important features. Through feature fusion, a time-series feature sequence for each electricity meter is generated. This sequence comprehensively describes the operating characteristics of the electricity meter at different time scales and is a high-dimensional description of the overall state of the electricity meter.

[0028] Cluster analysis was performed on the time-series feature sequences of each electricity meter. Clustering algorithms such as K-means and DBSCAN were used to group the time-series feature sequences based on similarity, classifying similar states into the same category. Through clustering, the operating states of the electricity meters were divided into multiple categories, each representing an operating mode of the meter, such as normal operation, load fluctuation, and fault warning. The clustering results generated multiple sets of operating state categories, each containing a group of similar electricity meter states.

[0029] The clustered set of operating status categories is associated with the time-series feature sequences of the electricity meters. The time-series feature sequence of each electricity meter is mapped to the corresponding operating status category. The state space of each electricity meter is composed of its time-series feature sequence and the corresponding state category. The state space of the electricity meter represents all operating statuses of the electricity meter within a specific time period and provides a basis for subsequent fault prediction and diagnosis.

[0030] Furthermore, anomaly gradient detection is performed on the state spaces of the multiple energy meters to obtain multiple energy meter anomaly gradient matrices, including: For each energy meter's state space, the direction of state change is identified to obtain multiple sets of state change directions; the rate of state change is identified for each energy meter's state space to obtain multiple sets of state change rates; the amplitude of state change is identified for each energy meter's state space to obtain multiple sets of state change amplitudes; anomaly identification and time-series correlation analysis are performed on the multiple sets of state change directions, the multiple sets of state change rates, and the multiple sets of state change amplitudes to generate the multiple energy meter anomaly gradient matrix.

[0031] Each electricity meter's state space contains its operating states over different time periods, representing the temporal changes in the meter's time-series characteristics. By analyzing the trends in the meter's state space over a period of time, the direction of state changes can be identified. For example, a continuous increase (positive change) or decrease (negative change) in the meter's current over a certain period are both directions of state change. By analyzing the direction of change in the meter's state space, a set of state change directions is generated for each meter, containing information on the direction of state changes over different time periods, which helps to further understand the meter's operating trends.

[0032] The rate of state change describes how quickly the state of an electricity meter changes. By calculating the change in state value within each unit of time, the rate of state change of the electricity meter is identified. For example, if the voltage or current of the electricity meter changes drastically in a short period of time, the rate of change is high; if the change is slow, the rate of change is low. After rate identification of the state space of each electricity meter, multiple sets of state change rates are generated, which contain information on the rate of state change of the electricity meter in different time periods.

[0033] The amplitude of state change describes the absolute degree of state change of an electricity meter. By calculating the maximum change in the meter's state over a certain period of time, the amplitude of the state change is identified. For example, if the voltage suddenly changes from one value to another, the magnitude of the amplitude can help identify abnormal faults. By identifying the amplitude of the electricity meter's state change, multiple sets of state change amplitudes are generated, including the absolute amplitude of the meter's state change in various time periods, reflecting abnormal fluctuations or drifts in the meter.

[0034] Among the changes in the state of each electricity meter, including direction, rate, and amplitude, we identify which changes are within the normal range and which are abnormal. For example, if the current of an electricity meter changes too rapidly, exceeding the normal rate, or the voltage changes too drastically, it can be considered an abnormal change. By identifying these outliers, we can detect potential hidden faults. For each electricity meter, time-series correlation analysis is used to analyze the temporal relationship of state changes. Time-series analysis helps identify whether the changes in the electricity meter conform to the expected operating mode or whether there are trend changes, such as meter drift caused by the gradual aging of the electricity meter. By integrating the above anomaly identification and time-series correlation results, we generate multiple electricity meter anomaly gradient matrices. Each matrix represents the anomaly gradient information of an electricity meter in different time periods. The elements in the matrix reflect whether the direction, rate, and amplitude of the electricity meter's state changes exceed the normal range.

[0035] Furthermore, based on the latent fault mechanism library, multidimensional latent fault prediction is performed on the anomaly gradient matrix of each energy meter to obtain the fault prediction domain for each energy meter, including: The latent fault mechanism library includes a metering drift mechanism set, a component aging mechanism set, and a sampling distortion mechanism set. Based on the metering drift mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the metering drift path for each energy meter. Based on the component aging mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the component aging path for each energy meter. Based on the sampling distortion mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the sampling distortion path for each energy meter. The metering drift path, the component aging path, and the sampling distortion path of each energy meter are added to the fault prediction domain of each energy meter.

[0036] The latent fault mechanism library contains different types of fault mechanism sets, each describing the possible causes and manifestations of faults in the electricity meter. The metering drift mechanism set is related to the metering accuracy of the electricity meter, describing metering drift phenomena caused by factors such as equipment aging and environmental changes. These faults cause inaccurate meter readings, resulting in metering errors. The component aging mechanism set describes the performance degradation of internal components of the electricity meter, such as sensors and integrated circuits, over time. Component aging affects the measurement accuracy and stability of the electricity meter. The sampling distortion mechanism set is related to the data sampling process of the electricity meter, describing sampling errors caused by hardware or software problems. These distortions may cause the electricity meter to fail to correctly record or process measurement data, affecting accuracy.

[0037] Based on information from the metering drift mechanism set, the abnormal change matrix of the electricity meter is analyzed to determine whether metering drift problems are caused by environmental changes or equipment aging. Specifically, the analysis examines whether the operating data of the electricity meter shows a gradual trend and whether this change conforms to the characteristics of metering drift. Through the above analysis, a metering drift path is generated, which is the trajectory of potential metering drift in the future. The metering drift path is part of the fault prediction domain and can indicate the potential fault risks and changing trends of the electricity meter.

[0038] Based on the component aging mechanism set, the system analyzes whether the state changes of the electricity meter conform to the component aging pattern. For example, an increase in current or voltage fluctuations is due to the gradual aging of internal components of the electricity meter, leading to a decline in its performance. Historical data from the electricity meter is used to determine if there are any signs of aging. Based on the analysis of the component aging mechanism, a component aging path is generated for each electricity meter, describing the fault path experienced by the meter due to component aging. This path reflects the impact of internal component aging and helps identify potential fault types and severity.

[0039] Based on the sampling distortion mechanism set, the abnormal gradient matrix of the energy meter is analyzed to identify potential distortions during the sampling process. Sampling distortions are caused by sensor malfunctions, software defects, or external interference, which can lead to inaccurate energy meter measurements and even affect the accuracy of electricity metering. By analyzing the distortion of the sampled data, a sampling distortion path is generated for each energy meter. This path represents the possible trend of sampling distortion, indicating potential problems the equipment may encounter during the sampling process and providing information for fault prevention and repair.

[0040] The obtained paths are integrated, and the fault prediction domain of each electricity meter contains these three paths. These paths represent the types of latent faults that the electricity meter may experience in the future, such as metering drift, component aging, and sampling distortion. The fault prediction domain can accurately describe the possible fault types, fault development trends, and their probability of occurrence of the electricity meter, which provides complete data support for subsequent fault prediction, maintenance decisions, and prevention strategies.

[0041] Furthermore, based on the metering drift mechanism set, implicit fault prediction is performed on the abnormal gradual change matrix of each electricity meter to obtain the metering drift path of each electricity meter, including: Based on the metering drift mechanism set, correlation features are extracted from the abnormal gradient matrix of each energy meter to obtain multiple drift correlation feature sets; based on the multiple drift correlation feature sets, metering drift fault tree prediction is performed on the multiple energy meters to obtain multiple metering drift fault path sets; based on the credibility evaluation and screening of each metering drift fault path set, each drift path screening set is obtained; each drift path screening set is screened by minimizing the path length to generate the metering drift path for each energy meter.

[0042] The metering drift mechanism set contains various features related to electricity meter metering drift, and the anomaly gradient matrix describes the changes in the electricity meter's state during operation, particularly the changes in metering accuracy. Based on the metering drift mechanism set, features related to metering drift are extracted from the anomaly gradient matrix. These features include information such as the trend, fluctuation range, and drift amplitude of electricity metering changes. Each electricity meter corresponds to a drift-related feature set, containing all features related to its metering drift.

[0043] Fault tree analysis is used to assess the likelihood of different factors causing faults. In this step, a fault tree model is built using a drift-related feature set. By extrapolating different drift mechanisms and their possible combinations, the potential metering drift fault paths that the electricity meter may experience in the future are predicted. Based on historical data and existing feature sets, multiple possible paths leading to metering drift are simulated through logical relationships. Based on the fault tree prediction results, a set of metering drift fault paths is generated for each electricity meter. Each path set describes the various steps and changes that the electricity meter may experience during metering drift, reflecting the potential metering drift risk.

[0044] By evaluating the reliability of each metering drift fault path, the path most likely to cause an actual fault is selected. The reliability evaluation is based on factors such as historical fault data, drift characteristics, and equipment age; a higher reliability path indicates a greater probability of the corresponding fault occurring. Based on the reliability scores, the set of metering drift fault paths is filtered, retaining only those paths with high reliability to reduce false alarms and redundant information. A drift path filter set is generated for each energy meter, containing the paths most likely to cause metering drift.

[0045] The drift path selection set is filtered by minimizing path length. Path length refers to the number of steps or stages involved in the fault occurrence process. Generally, a shorter path indicates a higher probability of fault occurrence, as faults usually manifest earlier and have a more rapid impact. The shortest and most representative metering drift path is selected, while lengthy or low-probability paths are removed. This filtering process improves the efficiency and accuracy of fault prediction. After path length minimization filtering, the final metering drift path is generated for each energy meter.

[0046] Furthermore, state transition analysis based on time decay weights is performed on the fault prediction domain of each electricity meter to establish multiple implicit fault transition relationship models, including: The fault transfer direction is identified in the latent fault mechanism library to obtain multiple fault transfer pointers; based on the multiple fault transfer pointers and multiple electricity meter fault transfer event sets, fault transfer is deduced for the fault prediction domain of each electricity meter to obtain a fault transfer trend set for each electricity meter; the fault transfer trend set of each electricity meter is corrected according to the time decay weight to obtain a fault transfer correction set; the latent fault graph structure is represented for the fault prediction domain of each electricity meter according to the fault transfer correction set to obtain the multiple latent fault transfer relationship models.

[0047] By analyzing a database of latent fault mechanisms, we can identify possible fault transfer directions after a fault occurs in an energy meter. Fault transfer direction refers to the path along which a fault propagates within or between devices after it occurs. For example, if a component of the energy meter ages, the fault can transfer from that component to other interdependent components. The transfer direction can be sequential, i.e., from one component to another, or parallel, i.e., multiple components are affected simultaneously. These transfer directions serve as fault transfer pointers, indicating the path of fault propagation.

[0048] The fault transfer event set contains multiple fault transfer events, describing the fault transfer scenarios of different energy meters under different times and conditions. Based on the fault transfer pointer and the fault transfer event set, a fault prediction domain is extrapolated for each energy meter, simulating the propagation process of the fault along the time axis. Through this extrapolation, the fault propagation path, possible impact range, and evolution trend of the fault are predicted. The extrapolation results generate a fault transfer trend set, representing the fault transfer trend of each energy meter in the future time period.

[0049] The impact of fault transfer gradually weakens over time. Therefore, a time decay weight is introduced, meaning that the propagation impact of a fault gradually decreases with time. This decay weight is defined using exponential decay, linear decay, etc. The time decay weight is applied to the fault transfer trend set, adjusting the fault transfer trend of each electricity meter according to the decay rules to align with real-world time-related patterns. After time decay correction, a corrected fault transfer set is obtained, containing the fault propagation path after time decay, accurately reflecting the actual propagation of electricity meter faults in different time periods.

[0050] The latent fault transfer relationship model uses a graph structure to represent the fault propagation path. Each node in the graph represents an electricity meter, and the edges represent the fault propagation relationship between electricity meters. This graph structure visually displays the fault transfer relationships between electricity meters, helping to identify which meters might be affected after a fault occurs. By mapping the fault transfer correction set onto the graph structure, a latent fault transfer relationship model is constructed. This model reflects the fault propagation process of electricity meters over time and predicts the types of future faults and their impact range.

[0051] Furthermore, based on the group relationship set among the multiple energy meters, the group evolution effect compensation is applied to the multiple latent fault transfer relationship models to construct an energy meter group state evolution model, including: Based on the group relationship set, metering drift group evolution is performed on the multiple latent fault transfer relationship models to obtain the first effect characteristic of group evolution; based on the group relationship set, component aging group evolution is performed on the multiple latent fault transfer relationship models to obtain the second effect characteristic of group evolution; based on the group relationship set, sampling distortion group evolution is performed on the multiple latent fault transfer relationship models to obtain the third effect characteristic of group evolution; based on the first effect characteristic, the second effect characteristic, and the third effect characteristic of group evolution, global compensation is performed on the multiple latent fault transfer relationship models to generate the energy meter group state evolution model.

[0052] The group relationship set includes the interrelationships among multiple energy meters, such as spatial topological relationships, historical fault relationships, and state change synchronization relationships. Based on the group relationship set, a group evolution of metering drift is performed on multiple latent fault transfer relationship models. This process analyzes the metering drift paths of multiple energy meters as group behavior, identifying the propagation patterns and evolution trends of metering drift faults within the group. During the group evolution of metering drift, group effect characteristics related to metering drift are extracted. These characteristics reflect the overall changing trends of energy meters in the group due to metering drift.

[0053] By analyzing the population relationship set, the aging trends of internal components in multiple energy meters were identified, and a latent fault transfer relationship model was evolved based on these trends. Component aging in energy meters is a significant factor leading to long-term metering drift and measurement errors. Through population evolution analysis of the component aging process, the effect characteristics related to component aging were extracted. These characteristics describe how component aging affects overall fault propagation and state changes within the energy meter population, helping to predict which components will age first in future equipment, leading to performance degradation or failure.

[0054] This study analyzes the impact of sampling distortion on electricity meters within a group based on population relation sets. Sampling distortion refers to errors in measurement values ​​during data acquisition caused by hardware or software issues in the electricity meters. This distortion can gradually amplify in some electricity meters and affect the entire system. By analyzing the population evolution of sampling distortion, the study identifies population effect characteristics related to sampling distortion. These characteristics help identify the overall failure risk caused by sampling errors within the population and predict which electricity meters are most susceptible to the effects of sampling distortion, leading to inaccurate data or measurement errors.

[0055] This method integrates and compensates for the population evolution effects stemming from metering drift, component aging, and sampling distortion. The compensation process modifies the implicit fault transfer relationship model of each meter based on its actual condition. By considering the mutual influence of different effects within the population, this compensation improves the accuracy of fault prediction and better simulates the behavior patterns of the meter population. After global compensation, a population state evolution model is generated. This model reflects the collective behavior and state changes of all meters in the population, demonstrating the overall evolution trend of the meter population under the influence of multiple factors such as fault transfer, equipment aging, metering drift, and sampling distortion. This provides comprehensive guidance for preventative maintenance and risk management of equipment.

[0056] Furthermore, the multi-scale time series features include short-cycle fluctuation features, medium-cycle stability features, and long-cycle drift trend features.

[0057] Short-cycle fluctuation characteristics describe the fluctuations in an electricity meter over a short period, such as rapid changes in current and voltage. These changes are related to factors such as load fluctuations and transient interference, and are characterized by rapid and drastic variations. Medium-cycle stability characteristics primarily describe the relatively stable operating characteristics of the electricity meter under normal working conditions, corresponding to a longer time period and reflecting the behavior of the equipment under typical operating conditions. Long-cycle drift trend characteristics mainly reflect the performance change trend of the electricity meter over a long period, related to factors such as equipment aging and changes in metering accuracy. These characteristics are relatively slow but gradually emerge, revealing the gradual degradation of the equipment or the cumulative effect of metering errors.

[0058] Furthermore, the group relationship set includes spatial topological relationships, historical fault relationships, and state change synchronization relationships.

[0059] Spatial topology describes the physical distribution and interconnection of multiple electricity meters. This relationship reveals the layout of the meters in the physical network, such as whether they belong to the same power grid or are on the same power line. Historical fault relationships describe the correlation of past faults among multiple electricity meters. If some meters have similar fault records in the past or failed at the same time, there is a historical correlation between them. State change synchronization relationships describe whether the state changes of multiple electricity meters are synchronized within the same time period. If multiple electricity meters experience similar state changes at the same or similar times, it indicates that they are affected by the same factors.

[0060] Example 2 is based on the same inventive concept as the time-series feature-based clustering method for predicting latent faults in electricity meters in the previous examples, such as... Figure 2 As shown in the embodiment of this application, a latent fault prediction system for electricity meters based on time-series feature clustering is provided. The system includes: The time-series feature clustering module 10 is used to acquire multiple operational datasets of multiple energy meters within the target area, and to perform multi-scale time-series feature extraction and time-series feature clustering on the multiple operational datasets to establish multiple energy meter state spaces; the anomaly gradient detection module 20 is used to perform anomaly gradient detection on the multiple energy meter state spaces to acquire multiple energy meter anomaly gradient matrices; the latent fault prediction module 30 is used to perform multi-dimensional latent fault prediction on the anomaly gradient matrix of each energy meter according to the latent fault mechanism library to acquire the fault prediction domain of each energy meter; the state transition analysis module 40 is used to perform state transition analysis based on time decay weight on the fault prediction domain of each energy meter to establish multiple latent fault transition relationship models; and the group evolution effect compensation module 50 is used to compensate for the group evolution effect of the multiple latent fault transition relationship models according to the group relationship set between the multiple energy meters to construct an energy meter group state evolution model.

[0061] Furthermore, the temporal feature clustering module 10 is used to perform the following operation steps: Each running dataset is segmented according to a preset time window to obtain a set of running segments for each electricity meter; features are extracted from the set of running segments for each electricity meter based on multi-scale time-series features to obtain a feature set for each electricity meter; features are fused from the feature sets for each electricity meter to generate a time-series feature sequence for each electricity meter; time-series feature clustering is performed on the time-series feature sequences for each electricity meter to obtain multiple sets of running state categories; and association mapping is performed on the time-series feature sequences for each electricity meter based on the multiple sets of running state categories to generate a state space for the multiple electricity meters.

[0062] Furthermore, the abnormal gradient detection module 20 is used to perform the following operation steps: For each energy meter's state space, the direction of state change is identified to obtain multiple sets of state change directions; the rate of state change is identified for each energy meter's state space to obtain multiple sets of state change rates; the amplitude of state change is identified for each energy meter's state space to obtain multiple sets of state change amplitudes; anomaly identification and time-series correlation analysis are performed on the multiple sets of state change directions, the multiple sets of state change rates, and the multiple sets of state change amplitudes to generate the multiple energy meter anomaly gradient matrix.

[0063] Furthermore, the latent fault prediction module 30 is used to perform the following operation steps: The latent fault mechanism library includes a metering drift mechanism set, a component aging mechanism set, and a sampling distortion mechanism set. Based on the metering drift mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the metering drift path for each energy meter. Based on the component aging mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the component aging path for each energy meter. Based on the sampling distortion mechanism set, latent fault prediction is performed on the abnormal gradient matrix of each energy meter to obtain the sampling distortion path for each energy meter. The metering drift path, the component aging path, and the sampling distortion path of each energy meter are added to the fault prediction domain of each energy meter.

[0064] Furthermore, the latent fault prediction module 30 is used to perform the following operation steps: Based on the metering drift mechanism set, correlation features are extracted from the abnormal gradient matrix of each energy meter to obtain multiple drift correlation feature sets; based on the multiple drift correlation feature sets, metering drift fault tree prediction is performed on the multiple energy meters to obtain multiple metering drift fault path sets; based on the credibility evaluation and screening of each metering drift fault path set, each drift path screening set is obtained; each drift path screening set is screened by minimizing the path length to generate the metering drift path for each energy meter.

[0065] Furthermore, the state transition analysis module 40 is used to perform the following operation steps: The fault transfer direction is identified in the latent fault mechanism library to obtain multiple fault transfer pointers; based on the multiple fault transfer pointers and multiple electricity meter fault transfer event sets, fault transfer is deduced for the fault prediction domain of each electricity meter to obtain a fault transfer trend set for each electricity meter; the fault transfer trend set of each electricity meter is corrected according to the time decay weight to obtain a fault transfer correction set; the latent fault graph structure is represented for the fault prediction domain of each electricity meter according to the fault transfer correction set to obtain the multiple latent fault transfer relationship models.

[0066] Furthermore, the population evolution effect compensation module 50 is used to perform the following operation steps: Based on the group relationship set, metering drift group evolution is performed on the multiple latent fault transfer relationship models to obtain the first effect characteristic of group evolution; based on the group relationship set, component aging group evolution is performed on the multiple latent fault transfer relationship models to obtain the second effect characteristic of group evolution; based on the group relationship set, sampling distortion group evolution is performed on the multiple latent fault transfer relationship models to obtain the third effect characteristic of group evolution; based on the first effect characteristic, the second effect characteristic, and the third effect characteristic of group evolution, global compensation is performed on the multiple latent fault transfer relationship models to generate the energy meter group state evolution model.

[0067] Furthermore, the multi-scale time series features include short-cycle fluctuation features, medium-cycle stability features, and long-cycle drift trend features.

[0068] Furthermore, the group relationship set includes spatial topological relationships, historical fault relationships, and state change synchronization relationships.

[0069] Through the foregoing detailed description of the method for predicting latent faults in electricity meters based on time-series feature clustering, those skilled in the art can clearly understand the latent fault prediction system for electricity meters based on time-series feature clustering in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting latent faults in electricity meters based on time-series feature clustering, characterized in that, The method includes: Multiple operational datasets of multiple energy meters within the target area are obtained, and multi-scale time-series feature extraction and time-series feature clustering are performed on the multiple operational datasets to establish multiple energy meter state spaces. Anomaly gradient detection is performed on the state space of the multiple energy meters to obtain multiple energy meter anomaly gradient matrices; Based on the hidden fault mechanism library, multidimensional hidden fault prediction is performed on the abnormal gradual change matrix of each energy meter to obtain the fault prediction domain of each energy meter. A state transition analysis based on time decay weight is performed on the fault prediction domain of each energy meter to establish multiple implicit fault transition relationship models. Based on the group relationship set among the multiple energy meters, the group evolution effect compensation is performed on the multiple latent fault transfer relationship models to construct an energy meter group state evolution model.

2. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, Multi-scale time-series feature extraction and time-series feature clustering are performed on the multiple running datasets to establish multiple energy meter state spaces, including: Each running dataset is segmented according to a preset time window to obtain a set of running segments for each electricity meter. Based on multi-scale time-series features, feature extraction is performed on the set of operating segments of each energy meter to obtain the feature set of each energy meter; The feature sets of each energy meter are fused to generate a time-series feature sequence for each energy meter. Perform time-series feature clustering on the time-series feature sequences of each energy meter to obtain multiple sets of operating status categories; The time-series feature sequences of each energy meter are associated and mapped according to the multiple sets of operating state categories to generate the state space of the multiple energy meters.

3. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, Anomaly gradient detection is performed on the state spaces of the multiple energy meters to obtain multiple energy meter anomaly gradient matrices, including: The state change direction is identified in the state space of each electricity meter to obtain multiple sets of state change directions; The state change rate of each energy meter is identified to obtain multiple sets of state change rates. For each energy meter, the state change amplitude is identified in the state space to obtain multiple sets of state change amplitudes; Anomaly identification and time-series correlation analysis are performed on the multiple sets of state change direction, the multiple sets of state change rate, and the multiple sets of state change amplitude to generate the multiple energy meter anomaly gradient matrix.

4. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, Based on the latent fault mechanism library, multidimensional latent fault prediction is performed on the anomaly gradient matrix of each energy meter to obtain the fault prediction domain for each energy meter, including: The hidden fault mechanism library includes a set of metering drift mechanisms, a set of component aging mechanisms, and a set of sampling distortion mechanisms; Based on the metering drift mechanism set, perform latent fault prediction on the abnormal gradual change matrix of each energy meter to obtain the metering drift path of each energy meter; Based on the set of component aging mechanisms, perform latent fault prediction on the abnormal gradual change matrix of each energy meter to obtain the aging path of each energy meter component; Based on the sampling distortion mechanism set, perform latent fault prediction on the abnormal gradient matrix of each energy meter to obtain the sampling distortion path of each energy meter; The metering drift path, the component aging path, and the sampling distortion path of each electricity meter are added to the fault prediction domain of each electricity meter.

5. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 4, characterized in that, Based on the metering drift mechanism set, implicit fault prediction is performed on the abnormal gradual change matrix of each electricity meter to obtain the metering drift path of each electricity meter, including: Based on the metering drift mechanism set, correlation features are extracted from the abnormal gradual change matrix of each energy meter to obtain multiple drift correlation feature sets; Based on the multiple drift-related feature sets, metering drift fault tree prediction is performed on the multiple energy meters to obtain multiple metering drift fault path sets; Based on the credibility evaluation of each set of meter drift fault paths, a set of drift path screenings is obtained. The path length of each drift path filter set is minimized to generate the metering drift path for each energy meter.

6. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, State transition analysis based on time decay weights is performed on the fault prediction domain of each electricity meter to establish multiple implicit fault transition relationship models, including: The fault transfer direction is identified in the hidden fault mechanism library to obtain multiple fault transfer pointers; Based on the multiple fault transfer pointers, and according to the multiple electricity meter fault transfer event sets, fault transfer inference is performed on the fault prediction domain of each electricity meter to obtain the fault transfer trend set of each electricity meter. Based on the time decay weight, the fault transfer trend set of each energy meter is corrected to obtain each fault transfer correction set; Based on the fault transfer correction sets, the fault prediction domains of each energy meter are characterized by a latent fault graph structure to obtain the multiple latent fault transfer relationship models.

7. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, Based on the group relationship set among the multiple energy meters, the group evolution effect compensation is performed on the multiple latent fault transfer relationship models to construct an energy meter group state evolution model, including: Based on the group relationship set, the multiple latent fault transfer relationship models are subjected to quantitative drift group evolution to obtain the first effect characteristics of group evolution; Based on the group relationship set, component aging group evolution is performed on the multiple latent fault transfer relationship models to obtain the second effect characteristics of group evolution; Based on the group relationship set, the multiple latent fault transfer relationship models are sampled and distorted for group evolution to obtain the third effect characteristics of group evolution; The multiple latent fault transfer relationship models are globally compensated based on the first effect characteristic, the second effect characteristic, and the third effect characteristic of population evolution to generate the energy meter population state evolution model.

8. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 2, characterized in that, The multi-scale time series features include short-cycle fluctuation features, medium-cycle stability features, and long-cycle drift trend features.

9. The method for predicting latent faults in electricity meters based on time-series feature clustering as described in claim 1, characterized in that, The group relationship set includes spatial topological relationships, historical fault relationships, and state change synchronization relationships.

10. A latent fault prediction system for electricity meters based on time-series feature clustering, characterized in that, The system is used to implement the method for predicting latent faults in electricity meters based on time-series feature clustering as described in any one of claims 1-9, the system comprising: The time-series feature clustering module is used to obtain multiple operational datasets of multiple energy meters within the target area, and to perform multi-scale time-series feature extraction and time-series feature clustering on the multiple operational datasets to establish multiple energy meter state spaces. An abnormal change detection module is used to detect abnormal changes in the state space of the multiple energy meters and obtain multiple abnormal change matrices of the energy meters. The latent fault prediction module is used to perform multi-dimensional latent fault prediction on the abnormal gradient matrix of each energy meter based on the latent fault mechanism library, and obtain the fault prediction domain of each energy meter. The state transition analysis module is used to perform state transition analysis based on time decay weight on the fault prediction domain of each energy meter and establish multiple implicit fault transition relationship models. The group evolution effect compensation module is used to compensate for the group evolution effect of the multiple latent fault transfer relationship models based on the group relationship set among the multiple energy meters, and to construct the energy meter group state evolution model.

Images