A knowledge graph construction and analysis system for smart meter data

Abstract

The application discloses a kind of knowledge graph construction and analysis system of intelligent electric meter data, including following unit: knowledge graph construction unit, abnormality detection unit, graph reasoning unit, diagnosis report generation unit and intelligent alarm unit;Through knowledge graph construction unit, the multidimensional electric power data of intelligent electric meter is acquired and preprocessed, then based on the data after pre-processing, the electric power field knowledge graph of associated electric meter equipment information, electric power parameter data, fault record and equipment association is constructed, and the application relates to electric power data processing and analysis technical field.The knowledge graph construction and analysis system of intelligent electric meter data, by constructing the knowledge graph of electric power field, system can comprehensively preprocess and correlation analysis to the multidimensional electric power data of intelligent electric meter, accurately capture the relationship between electric power parameter, equipment information and fault record, improve the precision of abnormality detection and fault reasoning.

Description

Technical Field

[0001] This invention relates to the field of power data processing and analysis, and more specifically, to a knowledge graph construction and analysis system for smart meter data. Background Technology

[0002] With the widespread adoption of smart meters in power systems, real-time data collection and analysis have become crucial for improving grid efficiency and ensuring power supply security. Smart meters record real-time power data, including parameters such as current, voltage, and power, and upload this data to power companies for analysis via communication networks. However, with the rapid increase in power data volume, traditional data analysis methods face numerous challenges, such as insufficient accuracy in detecting data anomalies, complexity in reasoning about fault causes, and untimely alarm processing.

[0003] Existing smart meter data analysis systems mostly focus on single data processing or anomaly detection algorithms, lacking in-depth mining and integration of complex relationships between power data. This leads to frequent missed or false alarms during large-scale power data monitoring, and makes it difficult to diagnose fault causes in a timely manner. Especially when dealing with complex faults and anomalies in power systems, current technologies have not yet provided sufficiently intelligent solutions, making it difficult to meet the needs of smart grids for efficient and accurate fault diagnosis and handling.

[0004] Therefore, building a system based on knowledge graph technology that can comprehensively process multi-dimensional data from smart meters, automatically detect anomalies, deduce the causes of faults, and generate diagnostic reports has become an urgent need to improve the level of smart meter data analysis and fault management. Summary of the Invention

[0005] The purpose of this invention is to provide a knowledge graph construction and analysis system for smart meter data. This system solves the problem that most existing smart meter data analysis systems focus on single data processing or anomaly detection algorithms, lacking in-depth mining and integration of complex relationships between power data. This leads to missed or false alarms during large-scale power data monitoring, and makes it difficult to diagnose fault causes in a timely manner, thus failing to meet usage requirements.

[0006] This invention achieves the above objectives through the following technical solution: a knowledge graph construction and analysis system for smart meter data, comprising the following units: The system includes a knowledge graph construction unit, an anomaly detection unit, a graph reasoning unit, a diagnostic report generation unit, and an intelligent alarm unit. The knowledge graph construction unit acquires and preprocesses multi-dimensional power data from smart meters, and then constructs a power domain knowledge graph based on the preprocessed data, which includes information on related meter devices, power parameter data, fault records, and device relationships. The anomaly detection unit integrates machine learning algorithms and knowledge graph features to monitor abnormal fluctuations in power data. The graph reasoning unit combines knowledge graph relationships, reasoning mechanisms, and historical data to infer possible causes of failures corresponding to anomalies. The diagnostic report generation unit automatically generates a diagnostic report containing details of the anomaly, the cause of the fault, and handling suggestions; The intelligent alarm unit issues corresponding intelligent alarm signals based on the level of abnormality, thereby realizing power anomaly detection and fault management.

[0007] Furthermore, the multi-dimensional power data acquired by the knowledge graph construction unit includes: The data must be at least one of the following: current data, voltage data, power data, electricity usage duration data, and meter identification information, and the data type must be time-series structured data.

[0008] Furthermore, the knowledge graph construction unit preprocesses the multi-dimensional power data, including: Data cleaning and standardization; Data cleaning is used to remove missing values, outliers, and invalid data caused by data acquisition errors. Standardization is used to unify the data dimensions of different power parameters.

[0009] Furthermore, the knowledge graph in the power sector includes core entity types and relationships between entities; The core entity types include at least one of the following: smart meter device entity, power parameter entity, fault type entity, device associated entity, and time entity; The relationships between entities include at least one of the following: collection relationship, connection relationship, trigger relationship, and correspondence relationship.

[0010] Furthermore, the knowledge graph construction unit uses a graph database to store the knowledge graph data; The preprocessed power data is mapped to the attribute values ​​of the corresponding entities in the knowledge graph; Simultaneously, the association weights between entities are calculated to improve the knowledge graph.

[0011] Furthermore, the machine learning algorithm fused by the anomaly detection unit includes the following steps: Models based on ensemble learning and time-series anomaly detection models; The anomaly detection unit extracts entity association features, power parameter time-series features, and historical fault association features from the knowledge graph to form a multi-dimensional feature set; It is then fed into an anomaly detection model built using a machine learning algorithm; Calculate the anomaly score and compare it with a preset threshold; Determine if there are any abnormal fluctuations in power data.

[0012] Furthermore, when the graph inference unit infers the possible causes of the fault corresponding to the anomaly, it includes the following steps: Construct a reasoning rule base that includes mapping rules between abnormal power parameters and fault types, rules for equipment association and fault propagation, and rules for association of historical fault recurrence; Employ at least one of the following reasoning mechanisms: path reasoning, semantic reasoning, and probabilistic reasoning; By combining historical fault handling data with real-time anomaly characteristics, the causes of faults can be inferred.

[0013] Furthermore, the diagnostic report generation unit generates a diagnostic report, including the following steps: Extract key information from the anomaly detection results, including the time of anomaly occurrence, the type of abnormal power parameter, the range of abnormal values, and the confidence level of the anomaly. A list of fault causes output by the association graph inference unit; Match the corresponding processing suggestions in the fault handling solution library of the knowledge graph; A structured diagnostic report is generated by integrating information according to a preset report template.

[0014] Furthermore, the intelligent alarm unit determines the anomaly level by including the following steps: Anomaly level score is calculated based on the anomaly confidence level and the severity coefficient corresponding to the cause of the fault in the anomaly detection results. Based on the score, the abnormality is classified into at least one of the following categories: minor abnormality, general abnormality, and severe abnormality; Configure corresponding alarm methods for different anomaly levels.

[0015] Furthermore, the alarm methods include: At least one of the following: system pop-up prompts, SMS notifications, email alerts, and audible and visual alarms; Severe anomalies trigger simultaneous alarms across multiple channels. The alarm signals include access to the diagnostic report and a summary of the core anomaly information, enabling quick location of the anomaly and viewing of diagnostic details.

[0016] The beneficial effects of this invention are as follows: 1. By constructing a knowledge graph in the power field, the system can perform comprehensive preprocessing and correlation analysis on multi-dimensional power data from smart meters, accurately capture the relationships between power parameters, equipment information, and fault records, and improve the accuracy of anomaly detection and fault reasoning.

[0017] 2. By combining machine learning algorithms with the multi-dimensional features of knowledge graphs, the system can automatically detect abnormal fluctuations in power data and infer possible causes of faults through a graph-based reasoning mechanism. This intelligent detection and reasoning approach effectively reduces human intervention and improves the accuracy and response speed of fault diagnosis.

[0018] 3. Based on the detected abnormal data and the inferred causes of the faults, a structured diagnostic report is automatically generated. The report includes details of the abnormality, the cause of the fault, and handling suggestions, greatly simplifying the work of power operation and maintenance personnel and improving work efficiency.

[0019] 4. The system, through its intelligent alarm unit, can automatically issue alarm signals based on the level of abnormality, ensuring that relevant personnel are promptly notified when power equipment malfunctions. Depending on the severity of the abnormality, the system provides multiple alarm methods, including pop-up notifications, SMS alerts, email reminders, and audible and visual alarms, ensuring timely fault handling.

[0020] 5. By realizing intelligent anomaly detection and fault management, the system can effectively reduce the failure rate of power equipment and improve the safety and stability of the power system through timely anomaly warning and fault reasoning.

[0021] 6. Based on historical fault data and real-time monitoring data, the system continuously optimizes inference rules and anomaly detection models, possesses self-learning and adaptability, and can cope with the ever-changing operating environment and complex fault situations in the power system. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a system block diagram of the present invention; Figure 2 This is a flowchart of the knowledge graph construction process for this invention; Figure 3 This is a flowchart of the anomaly detection and alarm process of the present invention. Detailed Implementation

[0023] The present application will now be described in further detail with reference to the accompanying drawings. It should be noted that the following specific embodiments are only used to further illustrate the present application and should not be construed as limiting the scope of protection of the present application. Those skilled in the art can make some non-essential improvements and adjustments to the present application based on the above application content. Example

[0024] Please see Figure 1-3This invention provides a technical solution: a knowledge graph construction and analysis system for smart meter data. The system comprises a knowledge graph construction unit, an anomaly detection unit, a graph reasoning unit, a diagnostic report generation unit, and an intelligent alarm unit. The system achieves power anomaly detection and fault management through the fusion of knowledge graphs and machine learning algorithms. The specific process includes: Acquire multi-dimensional power data from smart meters and preprocess it; Among them, the multi-dimensional power data of smart meters refers to the power-related information collected by smart meters involving multiple aspects, such as data in different dimensions such as voltage, current, power, and power factor; preprocessing is to clean, transform, and normalize the acquired raw power data to eliminate problems such as noise, missing values, and outliers in the data, so that the data meets the requirements of subsequent analysis. A knowledge graph for the power sector is constructed based on preprocessed power data, linking meter equipment information, power parameter data, historical fault records, and equipment relationships. The power sector knowledge graph is built upon relevant concepts, entities, and relationships within the power sector. It represents information such as smart meter equipment information, power parameter data, historical fault records, and equipment relationships. Smart meter equipment information includes various details about the smart meter itself, such as its model, manufacturer, installation location, and years of operation. Power parameter data refers to the various power parameters measured by the smart meter, such as the specific numerical values ​​of voltage and current mentioned above. Historical fault records are detailed records of faults that occurred in smart meters and related power equipment over a past period, including the time of occurrence, the symptoms, and the handling methods. Equipment relationships represent the various connections between smart meters and other power equipment, such as the connection between a particular meter and a specific transformer or line. By integrating machine learning algorithms and knowledge graph features through an anomaly detection unit, abnormal fluctuations in power data can be monitored in real time. Among them, machine learning algorithms are a class of algorithms that enable computers to automatically learn patterns and rules from data, thereby predicting and classifying new data. In anomaly detection, these algorithms can be used to identify abnormal patterns in power data. Knowledge graph features are the various entity attributes, relationships, and other information features contained in a knowledge graph. These features can provide additional contextual information for machine learning algorithms, assisting in anomaly detection. Abnormal fluctuations are changes in power data that deviate from the normal range or expected patterns, which may indicate the existence of power anomalies. By utilizing the graph reasoning unit based on the association and reasoning mechanism of the knowledge graph, and combining historical data, the possible causes of failures corresponding to anomalies are inferred. Among them, the relationship is the various connections between entities in the knowledge graph, such as the causal relationship between electricity meters and faults, and the connection relationship between devices; the reasoning mechanism is the process of deriving unknown information from known information based on the structure and rules of the knowledge graph through logical reasoning, similarity calculation and other methods; historical data is the electricity-related data recorded in the past, including power parameter data, fault records, etc., which is used to assist in reasoning the possible causes of faults corresponding to anomalies. The diagnostic report generation unit automatically generates a diagnostic report containing details of the anomaly, the cause of the failure, and handling suggestions. The anomaly details are a detailed description of the detected power anomaly, including the time and location of the anomaly, the metering equipment involved, and the specific manifestations of the anomaly; the cause of the fault is the possible cause of the power anomaly derived through graph inference; the handling suggestions are the corresponding solutions or suggestions given for the detected anomaly and the inferred cause of the fault, such as checking specific equipment or adjusting power parameters; The intelligent alarm unit issues a corresponding intelligent alarm signal based on the level of abnormality. Among them, the abnormality level is divided into different levels according to factors such as the severity and scope of the power abnormality; the intelligent alarm signal is a signal with specific meaning issued according to different abnormality levels, which can be in the form of sound signal, light signal, SMS notification, system pop-up window, etc., to remind relevant personnel to deal with abnormal situations in a timely manner.

[0025] It should be noted that, in use, the knowledge graph construction unit can integrate multi-dimensional data, linking information such as meter equipment, parameters, and fault records to form a comprehensive and structured knowledge system, providing a solid foundation for subsequent analysis. The anomaly detection unit integrates machine learning algorithms and knowledge graph features, enabling real-time and accurate monitoring of abnormal fluctuations, improving the accuracy and timeliness of detection. The graph reasoning unit, through association and reasoning mechanisms, combined with historical data, can deeply explore the causes of faults behind anomalies, enhancing fault analysis capabilities. The diagnostic report generation unit automatically generates reports containing key information, facilitating staff to quickly understand the situation. The intelligent alarm unit issues corresponding signals based on the anomaly level, promptly reminding relevant personnel to handle the situation and improving fault response speed. The overall design realizes intelligent and precise power anomaly detection and fault management, helping to ensure the stable operation of the power system.

[0026] In one embodiment, acquiring multi-dimensional power data from a smart meter and performing preprocessing includes: Acquire multi-dimensional raw electricity data collected by smart meters, including: Current data, voltage data, power data, electricity usage duration data, and meter identification information; The data format is time-series structured data: , in Indicates the data collection time step, for each data point. Include Each power parameter dimension, namely ; The raw power data is cleaned to remove missing values, outliers, and invalid data caused by data acquisition errors. Missing value removal uses neighborhood interpolation verification. If the same parameter data is missing at any time step, that time series segment is directly removed. Outlier removal uses the interquartile range method to calculate the quartiles of the parameter data. , and interquartile range When the data value exceeds If the range is exceeded, it is identified as an isolated point and removed. The cleaned power data is standardized using the following expression: , in Indicates the first The power parameter at the first The raw data at each time step For the first The average of several electrical parameters For the first The standard deviation of each electrical parameter The time step of the cleaned data. , .

[0027] This design acquires and cleans multi-dimensional raw data collected by smart meters, removing missing values ​​and outliers, and then standardizes the data. Cleaning removes invalid data, ensuring data quality and preventing it from interfering with subsequent analysis. Standardization brings different parameter data to the same dimension, facilitating comparison and calculation and improving data usability. The specific methods used to handle missing values ​​and outliers are scientific and reasonable, maximizing the retention of effective information and laying a solid foundation for subsequent operations such as building accurate knowledge graphs and precise anomaly detection.

[0028] In one embodiment, a knowledge graph in the power domain is constructed based on preprocessed power data, including: Define the core entity types of the knowledge graph, including smart meter device entities, power parameter entities, fault type entities, device association entities, and time entities; Establish relationships between entities, including the relationship between electricity meter devices and the collection of power parameters, the connection relationship between electricity meter devices and associated entities, the triggering relationship between power parameters and fault types, and the correspondence between fault types and historical records; The knowledge graph data is stored using a graph database. The preprocessed power data is mapped to entity attribute values ​​in the knowledge graph, and the association weights between entities are calculated. , in For entities With entity The frequency of association, To be with the entity The total number of related entities forms a complete knowledge graph of the power field, which includes nodes, edges, attributes, and association weights. Nodes represent various types of entities, edges represent the relationships between entities, and attributes include information such as specific values ​​of power parameters, equipment models, and fault occurrence times.

[0029] This design defines core entity types, builds relationships, uses a graph database to store and calculate association weights, clearly presents various elements in the power sector and their relationships, structures the scattered power data, facilitates efficient querying and management through graph database storage, and calculates association weights to reflect the closeness of the connections between entities. The complete knowledge graph provides comprehensive and accurate information for anomaly detection and reasoning, enabling the system to conduct in-depth analysis based on rich knowledge and improve the accuracy of fault diagnosis and handling.

[0030] In one embodiment, an anomaly detection unit integrates machine learning algorithms and knowledge graph features to monitor abnormal fluctuations in power data in real time, including: Extract entity association features, power parameter time-series features, and historical fault association features from the knowledge graph, and calculate the trend values ​​of the time-series features: , in A multi-dimensional feature set is formed by combining entity association weights with a preset time window step. , Historical fault association characteristics; An anomaly detection model integrating machine learning algorithms is constructed, using a multi-dimensional feature set as input. The machine learning algorithms include an ensemble learning model based on XGBoost and an LSTM-AE model for temporal anomaly detection. The loss function of the LSTM-AE model is: , in The power parameter data for model reconstruction, The objective function of the XGBoost model is: , For loss items, It is a regular term; Anomaly scores are calculated using an anomaly detection model. , in, These are the weighting coefficients. Let be the reconstruction loss of the LSTM-AE model at time t. Let be the anomaly probability output by the XGBoost model at time t; Determining the rules involves the following steps: Two key evaluation metrics were extracted from the knowledge graph: parameter temporal stability coefficient. Correlation strength coefficient with entity ; in, Through the formula: , The calculation is performed, where M is the number of parameters and T is the length of the time series. The smaller the value, the more stable the time series. The average weight of the edges connecting entities in the knowledge graph is used; the larger the value, the stronger the connection. The result was obtained through normalization. The final value: , in , This represents the extreme value of the time series stability coefficient. , This represents the extreme value of the correlation strength coefficient. The domain experience weight is set to 0.6, which is the optimal value verified by 1000 sets of historical power data. Specific applicable scenarios include: voltage parameter timing stability coefficient. When it is far below the average of 0.2, Automatically adjusted to 0.7, enhancing the timing detection advantages of LSTM-AE; when the correlation strength coefficient between electricity meters and transformers in a certain area... When (0.5 above the average), The value was reduced to 0.3, highlighting XGBoost's ability to discover associated features; when hour, An abnormal score threshold is preset to determine abnormal fluctuations in power data; Abnormal fluctuations include: sudden current changes, sudden voltage rises and falls, abnormal power deviations, and abnormal fluctuations in electrical load. The model outputs anomaly detection results and anomaly confidence scores: , This is the highest outlier score in history; The training process of the anomaly detection model combines historical fault data in the knowledge graph as labeled samples, and improves the model accuracy through iterative optimization, surpassing the traditional anomaly detection method based on fixed thresholds.

[0031] This design extracts multiple features to construct a multi-dimensional feature set, integrates two machine learning models to calculate anomaly scores, and combines the advantages of different models. XGBoost is good at processing structured data, while LSTM-AE is suitable for time series data and can comprehensively capture anomaly features. Combined with knowledge graph features, it considers the correlation between power data. By calculating anomaly scores and confidence levels, it accurately determines anomalies, surpassing the traditional fixed threshold method, improving the accuracy and flexibility of anomaly detection, and timely detecting abnormal fluctuations in power data.

[0032] In one embodiment, the graph reasoning unit utilizes the association relationships and reasoning mechanisms of the knowledge graph, combined with historical data, to reason about possible fault causes corresponding to anomalies, including: Construct a graph reasoning rule base, which includes rules for mapping power parameter anomalies to fault types, rules for equipment association and fault propagation, and rules for association of historical fault recurrence. Based on entity association paths in the knowledge graph, a mechanism combining path reasoning and semantic reasoning is used to calculate the confidence score of path reasoning: , in For the set of edges in the associated path, Assign association weights to edges to trace associated entities and potential influencing factors corresponding to abnormal power data; By integrating historical fault handling data with real-time anomaly features, the posterior probability of each possible fault cause is calculated using a Bayesian probabilistic inference algorithm.

[0033] in, For fault The prior probability is calculated from the historical failure frequency. Abnormal In the fault The conditional probability of occurrence is used to sort the posterior probabilities and output a Top-N list of fault causes, where N is the preset number of fault causes to be displayed.

[0034] This design constructs a reasoning rule base, combining path and semantic reasoning, and using Bayesian algorithms to calculate posterior probabilities. The constructed rule base can systematically cover power fault reasoning rules. The combination of path and semantic reasoning can comprehensively trace abnormal related entities and potential factors. Bayesian algorithms can integrate historical and real-time data, scientifically calculate the posterior probability of each fault cause, and output the results after sorting, providing a reliable basis for fault diagnosis, helping operation and maintenance personnel to quickly locate the root cause of the fault, and improving fault handling efficiency.

[0035] In one embodiment, a diagnostic report containing anomaly details, fault causes, and handling suggestions is automatically generated by the diagnostic report generation unit, including: Extract key information from the anomaly detection results, including the time of anomaly occurrence, the type of abnormal power parameter, the range of abnormal values, and the confidence level of the anomaly. ; The list of fault causes output by the association graph reasoning unit is combined with the fault handling solution library stored in the knowledge graph to match the standard handling suggestions corresponding to each fault cause, and the matching degree of the handling suggestions is calculated:

[0036] in For fault and the solution semantic similarity, The maximum semantic similarity; Based on the preset report template, the system integrates anomaly details, fault cause analysis, posterior probability ranking, and handling suggestions to automatically generate a structured diagnostic report, which supports export in text format and viewing within the system.

[0037] This design extracts key information about anomalies, correlates fault causes with handling suggestions, and generates reports based on templates. It automatically integrates anomaly detection and inference results, quickly generates structured reports, extracts key information to help staff quickly understand the anomaly overview, correlates fault causes with handling suggestions to provide targeted solutions, and supports exporting and viewing reports for easy information sharing and archiving. It provides a comprehensive and standardized reference for power fault handling, improving the standardization and efficiency of operation and maintenance work.

[0038] In one embodiment, the intelligent alarm unit issues a corresponding intelligent alarm signal based on the anomaly level, including: Based on anomaly confidence Severity coefficient of the cause of failure , The anomaly level score is calculated according to the power operation and maintenance specifications.

[0039] according to The value is used to classify the abnormality level: when At times, it is a minor abnormality. when This is a general abnormality. when This is a serious abnormality; Alarm methods are configured for different levels of anomalies, including system pop-up prompts, SMS notifications, email alerts, and audible and visual alarms. Severe anomalies trigger simultaneous alarms from multiple channels. The alarm signal includes a link to a diagnostic report and a summary of the core abnormal information, enabling staff to quickly locate the abnormal situation and view detailed diagnostic content, thus achieving rapid response and handling of power faults.

[0040] This design calculates a level score based on anomaly confidence and severity coefficients, classifies levels, and configures alarm methods. The levels are scientifically divided according to the actual situation of the anomaly, with different alarm methods corresponding to different levels. This highlights the urgency of severe anomalies, and multi-channel synchronous alarms ensure that staff receive information in a timely manner. The alarm signal includes a link to the diagnostic report and a core summary, which facilitates staff to quickly locate and view detailed information, enabling rapid response and handling of power faults and ensuring the stable operation of the power system.

[0041] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0042] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A knowledge graph construction and analysis system for smart meter data, characterized in that, It includes a knowledge graph construction unit, an anomaly detection unit, a graph reasoning unit, a diagnostic report generation unit, and an intelligent alarm unit; The knowledge graph construction unit is used to acquire multi-dimensional power data collected by smart meters. The multi-dimensional power data includes current data, voltage data, power data, electricity consumption duration data, and meter device identification information, and the data type is time-series structured data. The multi-dimensional power data is preprocessed, and a knowledge graph in the power field is constructed based on the preprocessed multi-dimensional power data. The core entity types of the knowledge graph in the power field include smart meter device entities, power parameter entities, fault type entities, device association entities, and time entities. The relationships between entities include the acquisition relationship between the meter device and the power parameter, the connection relationship between the meter device and the device association entity, the triggering relationship between the power parameter and the fault type, and the correspondence between the fault type and the historical fault record. The anomaly detection unit is used to extract entity association features, power parameter time series features, and historical fault association features from the power domain knowledge graph to form a multi-dimensional feature set. The multi-dimensional feature set is then input into an XGBoost-based ensemble learning model and an LSTM-AE-based time series anomaly detection model to obtain the anomaly probability output by the XGBoost model and the reconstruction loss output by the LSTM-AE model. The anomaly detection unit is also used to determine the fusion weight α based on the parameter temporal stability coefficient γ and the entity association strength coefficient δ of the power parameters; the parameter temporal stability coefficient γ is determined according to the following formula: ； Where M represents the number of power parameters, and T represents the length of the time series. Let be the parameter value of the m-th power parameter at time t. The smaller the value of the parameter temporal stability coefficient γ, the more stable the temporal state of the power parameter. The entity association strength coefficient δ is the average weight value of the entity association edges associated with the target smart meter in the power domain knowledge graph. The fusion weight α is determined according to the following formula: ； in, and These are the maximum and minimum values ​​of the time series stability coefficients, respectively. and These are the maximum and minimum values ​​of the entity association strength coefficient, respectively, and ω is a preset domain experience weight, which is set to 0.

6. The reconstruction loss and the anomaly probability are fused according to the fusion weight α to obtain the anomaly score. The abnormal score Determined according to the following formula: ； in, Let be the reconstruction loss of the LSTM-AE model at time t. Let t be the anomaly probability output by the XGBoost model at time t; and let the anomaly score be... Compared with a preset abnormal score threshold, when the abnormal score When the score exceeds the preset abnormal score threshold, it is determined that there is abnormal fluctuation in power data. The abnormal fluctuation in power data includes at least one of the following: sudden current change, sudden voltage rise and fall, abnormal power deviation, and abnormal fluctuation in power load. The graph reasoning unit is used to combine the relationships, reasoning mechanisms and historical fault handling data in the power field knowledge graph to reason about the possible causes of faults corresponding to abnormal fluctuations in the power data. The diagnostic report generation unit is used to generate a diagnostic report that includes details of the anomaly, the cause of the fault, and handling suggestions; The intelligent alarm unit is used to issue corresponding intelligent alarm signals according to the level of abnormality, so as to realize the power abnormality detection and fault management of smart meters.

2. The knowledge graph construction and analysis system for smart meter data according to claim 1, characterized in that, The knowledge graph construction unit preprocesses the multi-dimensional power data, including data cleaning and standardization. The data cleaning is used to remove missing values, outliers, and invalid data caused by data acquisition errors. The standardization is used to unify the data dimensions of different power parameters.

3. The knowledge graph construction and analysis system for smart meter data according to claim 1, characterized in that, The knowledge graph construction unit uses a graph database to store knowledge graph data, maps the preprocessed multi-dimensional power data to the attribute values ​​of corresponding entities in the power domain knowledge graph, and calculates the association weights of the edges between entities to form a power domain knowledge graph containing nodes, edges, attributes, and association weights.

4. The knowledge graph construction and analysis system for smart meter data according to claim 1, characterized in that, The training process of the anomaly detection model uses historical fault data from the power field knowledge graph as labeled samples, and updates the model parameters through iterative optimization.

5. The knowledge graph construction and analysis system for smart meter data according to claim 1, characterized in that, The anomaly detection unit outputs anomaly detection results and anomaly confidence levels. Determined according to the following formula: ； in, To preset anomaly score thresholds, This is the highest outlier score in history.

6. The knowledge graph construction and analysis system for smart meter data according to claim 1, characterized in that, The graph reasoning unit constructs a graph reasoning rule base, which includes mapping rules between abnormal power parameters and fault types, rules for equipment association and fault propagation, and rules for historical fault recurrence association. Based on the entity association paths in the power domain knowledge graph, a mechanism combining path reasoning and semantic reasoning is adopted, and combined with historical fault processing data and real-time anomaly features, possible fault causes are obtained through probabilistic reasoning.

7. The knowledge graph construction and analysis system for smart meter data according to claim 5, characterized in that, The diagnostic report generation unit extracts the abnormal occurrence time, abnormal power parameter type, abnormal value range, and abnormal confidence level from the abnormal detection results, associates them with the fault cause list output by the graph reasoning unit, matches the corresponding processing suggestions in the fault handling scheme library in the power field knowledge graph, and generates a structured diagnostic report according to the preset report template.

8. The knowledge graph construction and analysis system for smart meter data according to claim 5, characterized in that, The intelligent alarm unit calculates an anomaly level score based on the anomaly confidence level and the severity coefficient corresponding to the cause of the fault. According to the anomaly level score, the anomaly is divided into minor anomaly, general anomaly or severe anomaly, and a corresponding alarm mode is configured for different anomaly levels.

9. The knowledge graph construction and analysis system for smart meter data according to claim 8, characterized in that, Configure corresponding alarm methods for different anomaly levels, including at least one of system pop-up prompts, SMS notifications, email alerts, and audible and visual alarms; A serious anomaly triggers at least two of the aforementioned alarm methods to simultaneously issue an alarm. The alarm signal includes a diagnostic report access point and a summary of the core anomaly information, which are used to locate the anomaly and view the diagnostic details.

Images