Transformer fault early warning system and method based on multi-source heterogeneous data fusion analysis

Abstract

The invention discloses a transformer fault early warning system and method based on multi-source heterogeneous data fusion analysis, and the system comprises a data collection module which is used for collecting data from various heterogeneous data sources in real time or quasi real time; the data center module is used for cleaning, fusing, standardizing, modeling and managing the acquired data; the intelligent analysis module is used for acquiring data from the data center module, performing deep mining, feature extraction, state evaluation and fault prediction, and outputting an early warning signal and a research and judgment conclusion; the application service module is used for packaging various service capabilities in a micro-service form, and various services are set to be capable of being flexibly called and combined; and the display interaction module is used for providing interaction entrances of various functions. Through a micro-service and configurable architecture, deep fusion and intelligent analysis of multi-source, multi-format and multi-mode heterogeneous data are realized, and accurate evaluation of the state of the transformer and early-stage and active early warning of faults are realized.

Description

Technical Field

[0001] This invention relates to the field of power equipment technology, and more specifically, to a transformer fault early warning system and method based on multi-source heterogeneous data fusion analysis. Background Technology

[0002] As a core piece of equipment in the power system, the operational reliability of transformers directly affects the safety and stability of the power grid. For a long time, transformer condition monitoring and fault early warning technologies have been a key focus of industry research. Existing techniques mainly rely on the analysis of single or a few condition parameters. While progress has been made in specific areas, their limitations are becoming increasingly apparent when facing the complex multi-physics coupling effects and potential complex fault modes within the transformer. These limitations mainly include: 1. Traditional and intelligent methods based on dissolved gas analysis in oil Dissolved gas analysis (DGA) technology, which emerged and established its fundamental position in the 1960s and 70s, remains the cornerstone of transformer fault diagnosis. This technology is based on the principle that transformer insulating oil decomposes under electrical and thermal faults, producing specific gases (such as H2, CH4, C2H2, C2H4, etc.). The fault type is inferred by analyzing the gas composition and concentration. Early technology relied primarily on laboratory chromatography, with monitoring cycles lasting several months. It heavily depended on maintenance personnel using empirical formulas such as the Rogers ratio and Doernenburg ratio for manual judgment, essentially representing "post-event analysis" rather than "online early warning." With increasing demands for power grid reliability, online DGA monitors emerged in the 1980s and 90s, achieving a leap from periodic sampling to continuous automatic monitoring, shortening the monitoring cycle to days or hours. However, initial equipment was expensive and lacked stability. In the 21st century, sensor technology has evolved towards miniaturization and high reliability, and DGA has become a standard feature in important transformers. Meanwhile, the introduction of big data and artificial intelligence technologies has fundamentally changed the data analysis paradigm. Machine learning algorithms such as support vector machines, random forests, and neural networks are widely used, enabling the system to learn from massive amounts of historical data and build complex predictive models, significantly improving diagnostic accuracy. Furthermore, DGA data is beginning to be analyzed in conjunction with parameters such as partial discharge and temperature, showing an initial trend towards integration. The development trajectory of this technology can be summarized as: from human experience-based judgment, to online monitoring, and now to intelligent diagnosis and preliminary multi-source fusion.

[0003] 2. Partial discharge monitoring technology based on high frequency and ultra-high frequency Partial discharge is the primary sign of transformer insulation degradation, and its monitoring technology is one of the most sensitive and direct means of assessing insulation condition. Early academic research clarified that the electromagnetic wave signals generated by partial discharge cover the high-frequency to ultra-high-frequency bands. In the 1970s and 80s, monitoring mainly relied on electrical pulse detection methods, which are susceptible to field interference, while ultra-high-frequency technology was still in the laboratory theoretical research stage. It wasn't until the 1990s and early 21st century, after the successful application of ultra-high-frequency technology in gas-insulated switchgear, that it began to be tested on transformers. Sensors installed on the tank walls capture nanosecond-level discharge pulses, exhibiting superior anti-interference capabilities compared to traditional methods. However, this technology faces challenges such as sensor installation, complex signal propagation paths, and defect type identification. In recent years, with the development of signal processing and artificial intelligence technologies, partial discharge monitoring has achieved a leap from "discharge detection" to "discharge identification and location." Phase-analysis-based partial discharge pattern recognition can form a discharge "fingerprint map" and automatically identify the discharge type through AI algorithms; using ultra-high-frequency sensor arrays can achieve centimeter-level precise location of the discharge source through signal arrival time difference. Furthermore, the combined application of ultra-high frequency radio waves with localization methods such as ultrasound and acoustic emission has further improved the reliability of diagnosis. The evolution of this technology has followed a path from laboratory principle verification to interference-resistant online monitoring, and now to intelligent pattern recognition and precise localization.

[0004] 3. Mechanical condition monitoring technology based on vibration and noise analysis Vibration and noise analysis technology has evolved from the traditional practice of maintenance personnel relying on experience to judge the operating status of transformers using listening rods, and has achieved digital and intelligent upgrades. In the 1990s and early 21st century, with the development of acceleration vibration sensors and digital signal processing technology, quantitative analysis of transformer vibration signals (mainly originating from core magnetostriction and winding electromagnetic forces) became possible. The research focus at this stage was on extracting characteristic frequencies related to the mechanical state of the core and windings from complex background noise. In recent years, this technology has gradually moved towards intelligence, shifting its focus from basic feature extraction to deeper state assessment and fault early warning. Its development can be summarized as from establishing vibration baselines to trend analysis, and now to the deep application of AI.

[0005] However, despite significant progress in the aforementioned individual technologies, existing transformer fault early warning schemes still have obvious shortcomings in practical engineering applications: First, traditional methods rely on single data source analysis and lack the ability to collaboratively perceive and deeply integrate heterogeneous data from multiple sources such as electrical, chemical, mechanical, and thermodynamic data, resulting in insufficient accuracy in identifying complex faults and early defects; Second, due to data heterogeneity (differences in sampling frequency, dimensions, and spatiotemporal scales), existing methods struggle to establish unified feature representation models, limiting them to simple data overlay or post-decision fusion, and failing to uncover cross-modal correlation features; Third, early warning systems based on static thresholds or traditional machine learning models have poor generalization capabilities, rely on manual experience adjustments, are difficult to adapt to dynamic changes in different transformer types and operating conditions, and have weak interpretability of early warning results, lacking the ability to track fault evolution and support operation and maintenance decisions. Summary of the Invention

[0006] The purpose of this invention is to address the problems of data silos, one-sided analysis, and delayed response in transformer fault early warning systems in ultra-high voltage substations in the existing technology. It provides a transformer fault early warning system and method based on multi-source heterogeneous data fusion analysis. Through a microservice-based and configurable architecture, it achieves deep fusion and intelligent analysis of multi-source, multi-format, and multi-modal heterogeneous data, enabling accurate assessment of transformer status and early, proactive fault warning.

[0007] To achieve the above objectives, embodiments of the present invention provide a transformer fault early warning system based on multi-source heterogeneous data fusion analysis, the system comprising: The data acquisition module is equipped with various data interfaces and adapters for real-time or near-real-time data acquisition from various heterogeneous data sources. The data platform module is used to clean, integrate, standardize, model, and manage the collected data. The intelligent analysis module is used to acquire data from the data platform module, perform in-depth mining, feature extraction, status assessment and fault prediction, and output early warning signals and judgment conclusions. The application service module is used to encapsulate various business capabilities in the form of microservices, including early warning services, reporting services, 3D services, and video services. Furthermore, these services are configured to be flexibly invoked and combined. The interactive module uses a micro-frontend architecture to build a visual interface, providing interactive entry points for various functions.

[0008] Preferably, the data platform module includes a data lake and a data warehouse, wherein the data lake is used to store raw data and the data warehouse is used to store cleaned and processed standard data.

[0009] Preferably, the intelligent analysis module incorporates various machine learning and data analysis algorithm models, including unsupervised learning models and supervised learning models; among them, Unsupervised learning models include clustering algorithms, which are used to cluster normal states and historical anomalous states to discover new anomalous patterns or to detect anomalies. Supervised learning models include classification models, regression models, and deep learning models. Classification models are used to train a multi-classification model to determine the type of failure; regression models are used to predict the remaining life of equipment or the future values ​​of key indicators; and deep learning models are used to perform deeper feature extraction and pattern recognition on time-series and image data using Long Short-Term Memory Networks (LSTM) and Convolutional Neural Networks (CNN).

[0010] On the other hand, the present invention provides a transformer fault early warning method based on multi-source heterogeneous data fusion analysis. This method uses the above-mentioned system to perform transformer fault early warning based on multi-source heterogeneous data fusion analysis.

[0011] Preferably, the method includes: Collect multi-source heterogeneous data to comprehensively acquire all data related to transformer status; during collection, for real-time data streams, use message queues for subscription and push; for business data in relational databases, use data synchronization tools or scheduled tasks for incremental extraction; for file data, use file transfer services or object storage interfaces for uploading and retrieval; for third-party systems, obtain data by calling their provided RESTful API interfaces. The collected multi-source heterogeneous data is cleaned and standardized preprocessed. The cleaned multi-source data is deeply integrated in the data platform, and high-value features for fault early warning are extracted. The data fusion is carried out by combining feature-level fusion and decision-level fusion, and time-domain feature extraction, frequency-domain feature extraction, spectral feature extraction and trend feature extraction are performed according to feature engineering. By leveraging the fused high-dimensional features, intelligent analysis is performed through a pre-trained machine learning model to assess the status and provide early warnings of faults, outputting a health score or fault probability for the device. The warning results are transformed into actionable operation and maintenance suggestions and presented to users through the interface.

[0012] Preferably, multi-source heterogeneous data is collected to comprehensively acquire all data related to the transformer's condition, including: SCADA system is used to collect real-time operating data of transformers, circuit breakers and instrument transformers; The content and growth rate of each characteristic gas were collected using an oil chromatography monitoring system (DGA). A partial discharge monitoring system was used to collect UHF and ultrasonic AE signals to generate phase-resolved partial discharge PRPS, phase-resolved partial discharge pulse PRPD, and φ-QN spectra. The vibration acceleration and noise spectrum of the transformer tank are collected by a vibration acoustic monitoring system to analyze mechanical defects such as loose core clamps and winding deformation. The grounding current values ​​of the transformer core and clamps are collected by monitoring the grounding current of the core to determine whether there is a multi-point grounding fault. The pressure, density, and trace moisture content of the gas chamber in GIS equipment are monitored using an SF6 gas monitoring system. The auxiliary control system collects data from fire protection, security, environment, and video surveillance for comprehensive environmental risk assessment and alarm linkage. Real-time meteorological data of the site area is obtained using a meteorological system to analyze the impact of the external environment on the thermal stability and insulation performance of the equipment. Static data such as equipment ledger information, test reports, maintenance history, and family defect records are obtained through the Production Management System (PMS) to provide historical basis for condition assessment. When a fault occurs, the fault recording system collects and parses the fault recording file to obtain the waveform, amplitude, phase, and protection action sequence of the fault current and voltage. The robot inspection system receives infrared thermal images and visible light images transmitted back by the inspection robot to identify defects such as overheating and abnormal appearance of the equipment.

[0013] Preferably, the preprocessing of cleaning and standardizing the collected multi-source heterogeneous data includes: For outliers, business rules and statistical methods are used to identify and handle abnormal fluctuations and singularities; For null values, methods such as deletion, filling with the mean / median, or filling based on machine learning algorithm prediction are used, depending on the importance of the data. Standardize the format, unify the timestamp format, numerical units, and enumeration value encoding; Remove duplicate data caused by network retransmission; For unstructured data, feature extraction is performed to convert it into structured labels; Time alignment is performed for data streams of different frequencies; Additionally, data with different dimensions are scaled to the same numerical range to eliminate the influence of dimensions between features.

[0014] Preferably, feature-level fusion includes concatenating low-level feature vectors from different data sources into a high-dimensional joint feature vector; decision-level fusion includes first performing independent analysis and preliminary decision-making on different data sources or models, and then fusing the decision results; time-domain feature extraction includes calculating statistical features such as mean, variance, peak value, and kurtosis; frequency-domain feature extraction includes performing Fast Fourier Transform (FFT) on vibration and acoustic signals to extract fundamental frequency, harmonic components, and frequency centroid; spectral feature extraction includes extracting skewness, steepness, cross-correlation coefficient, and discharge phase distribution from partial discharge PRPD spectra; and trend feature extraction includes calculating the short-term and long-term rates of change of key indicators and the moving window average value to capture the deterioration trend of the indicators.

[0015] Preferably, the assessment of status and early warning of faults includes: setting multi-level early warning thresholds, and automatically triggering an early warning of the corresponding level when the score exceeds the threshold, wherein the early warning is set to take into account both instantaneous value exceeding the limit and trend warning.

[0016] Preferably, the interface presented to the user includes: It adopts micro-frontend and configurable large screen technology, allowing users to customize the monitoring interface by dragging and dropping components; It offers a rich set of visualization components, including David's triangle and three-ratio plots for visualizing oil chromatography data analysis results, PRPD / PRPS spectrum display components for interactive analysis of partial discharge types, trend comparison curves that support trend comparison analysis of multiple parameters and time periods, and 3D model linkage with decision support wizards.

[0017] Through the above technical solutions, this invention constructs a spatiotemporal alignment mechanism for multi-source heterogeneous data, an adaptive fusion model driven by deep learning, and an interpretable fault evolution analysis framework, thereby achieving early and accurate fault warning and intelligent recommendation of operation and maintenance strategies, significantly improving the comprehensiveness of transformer status perception, the accuracy of diagnosis, and the practicality of early warning.

[0018] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a transformer fault early warning system based on multi-source heterogeneous data fusion analysis provided by the present invention; Figure 2 This is a flowchart illustrating the transformer fault early warning method based on multi-source heterogeneous data fusion analysis provided by the present invention. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0021] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0022] See Figure 1 This invention provides a transformer fault early warning system based on multi-source heterogeneous data fusion analysis, the system comprising: The data acquisition module is equipped with various data interfaces and adapters for real-time or near-real-time data acquisition from various heterogeneous data sources. The data platform module is used to clean, integrate, standardize, model, and manage the collected data. The intelligent analysis module is used to acquire data from the data platform module, perform in-depth mining, feature extraction, status assessment and fault prediction, and output early warning signals and judgment conclusions. The application service module is used to encapsulate various business capabilities in the form of microservices, including early warning services, reporting services, 3D services, and video services. Furthermore, these services are configured to be flexibly invoked and combined. The interactive module uses a micro-frontend architecture to build a visual interface, providing interactive entry points for various functions.

[0023] According to the above technical solution, the system provided by this invention adopts a layered and decoupled design concept. The overall architecture includes a data acquisition module, a data middle platform module, an intelligent analysis module, an application service module, and a display and interaction module. The data acquisition module, as the system's data input, is responsible for acquiring data from various heterogeneous data sources in real-time or near real-time. This layer deploys multiple data interfaces and adapters to handle different communication protocols and data formats. The data middle platform module is the core data hub of the system, including a data lake (for storing raw data) and a data warehouse (for storing cleaned and processed standard data). This module is responsible for data cleaning, fusion, standardization, modeling, and management, providing unified, high-quality data services for upper-layer applications. The intelligent analysis module is the system's brain, incorporating various machine learning and data analysis algorithm models. This module acquires data from the data middle platform, performs in-depth mining, feature extraction, status assessment, and fault prediction, and outputs early warning signals and judgment conclusions. The application service module encapsulates various business capabilities in the form of microservices, such as early warning services, reporting services, 3D services, and video services. These services can be flexibly invoked and combined to support diverse upper-layer business applications. The interactive module adopts a micro-frontend architecture to build a configurable, drag-and-drop visual interface. It provides users with interactive entry points for functions such as panoramic monitoring, intelligent early warning, event analysis, and decision support, and supports multiple terminals such as PCs and large screens.

[0024] Furthermore, to address the problems of data silos, partial analysis, and delayed response in existing UHV substation transformer fault early warning systems, this invention provides a transformer fault early warning method based on multi-source heterogeneous data fusion analysis. This method uses the aforementioned system for transformer fault early warning based on multi-source heterogeneous data fusion analysis. Specifically, as... Figure 2 As shown, the method includes: First, multi-source heterogeneous data is collected to comprehensively acquire all data related to the transformer's condition; the data sources include, but are not limited to: 1. SCADA system: Collects real-time operating data of main equipment such as transformers, circuit breakers, and instrument transformers, such as three-phase current, voltage, active / reactive power, oil temperature, winding temperature, tap position, etc. 2. Online monitoring system: Oil chromatography monitoring system (DGA): Collects the content and growth rate of characteristic gases such as hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), and carbon dioxide (CO2); 3. Partial Discharge Monitoring System: Acquires ultra-high frequency (UHF) and ultrasonic (AE) signals, and generates raw data such as PRPS (phase-resolved partial discharge), PRPD (phase-resolved partial discharge pulse) spectra, and φ-QN spectra; 4. Vibration and acoustic monitoring system: Collects data such as transformer tank vibration acceleration and noise spectrum to analyze mechanical defects such as loose core clamps and winding deformation; 5. Core grounding current monitoring: Collect the grounding current values ​​of the transformer core and clamps to determine whether there is a multi-point grounding fault; 6. SF6 gas monitoring: For GIS equipment, monitor the pressure, density, and trace moisture content of its gas chamber; 7. Auxiliary control system: Collects data from fire protection, security, environment (temperature and humidity, water immersion, smoke detection), video surveillance, etc., for comprehensive environmental risk assessment and alarm linkage; 8. Meteorological System: Real-time meteorological data of the site area, such as ambient temperature, humidity, wind speed, rainfall, and lightning location information, are obtained through the interface to analyze the impact of the external environment on the thermal stability and insulation performance of the equipment. 9. Production Management System (PMS): Acquires static data such as equipment ledger information, test reports, maintenance history, and family defect records to provide historical basis for condition assessment; 10. Fault Recording System: When a fault occurs, it collects and parses fault recording files to obtain the waveforms, amplitudes, phases, and protection action sequences of the fault current and voltage. 11. Robot Inspection System: Receives infrared thermal imaging images and visible light images transmitted back by the inspection robot, and is used to identify defects such as equipment overheating and abnormal appearance.

[0025] In this embodiment, during data collection: For real-time data streams, message queues (such as Kafka, MQTT) are used for subscription and push. For business data in relational databases, incremental extraction is performed using data synchronization tools (such as Canal, Debezium) or scheduled tasks. For file data (such as waveform recordings and images), upload and retrieval are performed using file transfer services (FTP, SFTP) or object storage (such as S3) interfaces; For third-party systems (such as meteorology and dispatching systems), data is obtained by calling their provided RESTful API interfaces.

[0026] Secondly, the collected multi-source heterogeneous data undergoes cleaning and standardization preprocessing, including: 1. Data Cleaning: Handling Null / Outlier Values: Identify and handle abnormal fluctuations and singularities using business rules and statistical methods (such as the 3σ principle and box plots); for null values, use methods such as deletion, filling with the mean / median, or filling based on machine learning algorithms to predict and fill, depending on the importance of the data. 2. Standardize the format: unify the timestamp format (e.g., UTC timestamp), numerical units (e.g., temperature is uniformly in degrees Celsius), and enumeration value encoding (e.g., alarm levels are uniformly "normal, attention, abnormal, severe"). 3. Data deduplication: Remove duplicate data caused by network retransmission or other reasons; 4. Data transformation and alignment: Unstructured data (such as fault reports and inspection photos) is subjected to feature extraction and converted into structured labels (such as "maximum temperature: 65.3℃" and "oil seepage detected"). Time alignment is performed on data streams of different frequencies. For example, SCADA data in milliseconds and oil chromatography data in minutes are aggregated through a time window and then correlated. 5. Data standardization / normalization: Scaling data of different dimensions to the same numerical range eliminates the influence of dimensions between features, preparing for subsequent fusion and model training.

[0027] Next, the cleaned multi-source data is deeply fused in the data platform, and high-value features for fault early warning are extracted; the data fusion is performed using a combination of feature-level fusion and decision-level fusion. Feature-level fusion: This involves concatenating low-level feature vectors from different data sources into a high-dimensional joint feature vector. For example, the content of characteristic gases in oil chromatography, load current, top oil temperature, and ambient temperature at a certain moment can be concatenated into a single feature vector to comprehensively characterize the overall state of the transformer at that moment. Decision-level fusion: Different data sources or models first perform independent analysis and preliminary decisions, and then the decision results are fused. For example, the oil chromatography analysis model independently judges "medium-temperature overheating", and the vibration analysis model independently judges "slight winding deformation". By combining the outputs of the two models, the final decision is "winding overheating with deformation", which improves the confidence of the diagnosis.

[0028] Simultaneously, feature extraction is performed in the time domain, frequency domain, spectral, and trend domains based on feature engineering. 1. Temporal feature extraction: Calculate statistical features such as mean, variance, peak value, and kurtosis; 2. Frequency domain feature extraction: Perform Fast Fourier Transform (FFT) on vibration and acoustic signals to extract features such as fundamental frequency, harmonic components, and frequency centroid; 3. Feature extraction: Extract hundreds of features from the partial discharge PRPD map, including skewness, burtosis, cross-correlation coefficient, and phase distribution of discharge quantity. 4. Trend feature extraction: Calculate the short-term and long-term change rates and sliding window averages of key indicators (such as acetylene content and total hydrocarbon content) to capture the deterioration trend of the indicators.

[0029] This constructs a multi-source heterogeneous data fusion engine with built-in functions such as data pattern mapping, entity recognition, and data quality verification. It can understand the semantics of data from different sources; for example, it can identify that "phase A current" in a SCADA system and "IA" in an online monitoring system are the same entity and automatically associate them. Simultaneously, it supports integrated stream and batch processing, enabling both real-time data stream fusion analysis and batch mining of massive historical data.

[0030] Then, utilizing the aforementioned fused high-dimensional features, intelligent analysis is performed through a pre-trained machine learning model to assess the device's status and provide early warnings of faults, outputting a device health score or fault probability; where, Model building includes: 1. Unsupervised learning model: Clustering algorithms (such as K-Means, DBSCAN) are used to cluster normal states and historical abnormal states to discover new abnormal patterns or to detect anomalies. 2. Supervised learning models: Classification model: A multi-classification model is trained using gradient boosting trees (such as XGBoost, LightGBM), random forests, support vector machines (SVM), etc., to determine the type of fault (such as discharge fault, overheating fault, mechanical fault). Regression model: used to predict the future value of a device's remaining useful life (RUL) or key metrics; Deep learning models: For time-series data (such as vibration signals, DGA sequences) and image data (such as infrared spectra, partial discharge spectra), models such as Long Short-Term Memory Network (LSTM) and Convolutional Neural Network (CNN) are used to perform deeper feature extraction and pattern recognition.

[0031] Status assessment and early warning include: Multi-level warning thresholds can be set (e.g., alert, abnormal, severe). When the score exceeds the threshold, the corresponding level of warning is automatically triggered. This warning is not only based on instantaneous value exceeding the limit, but also focuses on trend warnings. For example, even if the absolute value of acetylene does not exceed the standard, but its continuous 7-day growth rate exceeds 50%, the system will still trigger a trend warning.

[0032] In this embodiment, a knowledge graph with the transformer as the core entity is constructed, integrating knowledge such as equipment ledgers, defect records, test data, family defects, and technical standards.

[0033] When an alert is issued, the system uses a knowledge graph for reasoning. For example: "The current transformer model is XX. This model is known to have a defect of the XX family. The current acetylene growth trend is similar to historical case #1234. This case was ultimately confirmed as an XX fault. It is recommended to take XX measures." The case library continuously learns from new outcomes, enabling the digital accumulation and intelligent reuse of experience.

[0034] Finally, the early warning results are transformed into actionable operation and maintenance suggestions and presented to users through the interface. Event-based analysis includes: when an early warning occurs, it doesn't simply list alarm signals, but automatically associates and aggregates relevant SCADA alarms, online monitoring data, video footage, fault recordings, 3D model localization, and other information based on a predefined event rule engine, forming a complete anomaly event summary. This summary includes: event occurrence time, equipment name, event type (e.g., "abnormal acetylene growth trend in converter transformer"), event level, associated data (trend curves, graphs, real-time video links), preliminary analysis conclusions, and standardized handling suggestions based on a case library.

[0035] In terms of visual interaction, a micro-frontend (using a Qiankun-based micro-frontend architecture, separating different functions such as panoramic monitoring, device status, and intelligent alerts into independent micro-applications, supporting independent development, deployment, and upgrades by different teams) and configurable large-screen technology can be adopted, allowing users to customize the monitoring interface by dragging and dropping components (providing business card configuration management and theme panel management functions. Each micro-application exists in the form of a business card, and users can drag and drop the required business cards into different theme panels to form a customized monitoring desktop). Specifically, the provided visualization components include: David's triangle diagram and three-ratio diagram: used to visualize the results of oil chromatography data analysis.

[0036] PRPD / PRPS graph display component: used for interactive analysis of partial discharge types.

[0037] Trend comparison curve: Supports trend comparison analysis with multiple parameters and time periods.

[0038] 3D model linkage: Highlight faulty equipment in the 3D visualization model and access all relevant information about the equipment with one click.

[0039] Based on the aforementioned visualization components, a decision support wizard is provided to guide operations and maintenance personnel to confirm, inspect, and handle issues according to standardized procedures, and to record the entire process to form a closed-loop management system.

[0040] The following is a specific implementation method to illustrate the above-mentioned transformer fault early warning method based on multi-source heterogeneous data fusion analysis: S1: Scene and Initial State Target equipment: High-end Y / Y converter transformer of pole I in a certain UHV converter station (equipment ID: T-1001).

[0041] System status: The system provided by this invention is deployed on the provincial cloud platform to monitor all main devices of the station 24 / 7.

[0042] Initial state: All SCADA operating parameters (current, voltage, temperature) of the device are displayed normally, and there are no traditional SCADA alarms.

[0043] S2: Data Acquisition and Preprocessing 1. Parallel acquisition of multi-source data: Oil chromatography online monitoring (DGA) system: Data from the T-1001 transformer oil sample was collected, showing that the acetylene (C2H2) content was 0.8 μL / L (Note: 1 μL / L). Although it did not exceed the standard, the system recorded the value and started tracking in the background.

[0044] Partial Discharge Online Monitoring System: Acquires ultra-high frequency (UHF) signals and generates a PRPD (Partial Discharge Perimeter) spectrum after preliminary preprocessing. The spectrum shows concentrated discharge signals in both the positive and negative half-cycles of the power frequency phase, but the absolute value of the discharge does not exceed the preset alarm threshold.

[0045] SCADA system: The load current is 85% of the rated value, the oil temperature is 65℃, and the operating conditions are stable.

[0046] Other systems: The video surveillance display equipment appeared normal; the environmental monitoring data showed no unusual changes.

[0047] 2. Data cleaning and standardization: After receiving the above data, the data platform immediately performs data cleaning.

[0048] All data timestamps were standardized to Beijing time, and the oil chromatogram data units were standardized to μL / L.

[0049] The UHF signal is noise-reduced to remove interference pulses from the external environment.

[0050] S3: Data Fusion and Feature Extraction The data fusion engine correlates and aligns DGA data, partial discharge data, and SCADA load current data within the same time window.

[0051] Feature-level fusion: Constructing a fused feature vector, for example: [C2H2=0.8, C2H2_7d_trend=+120%, Load current=0.85pu, Discharge capacity=1200pC, PRPD_skewness=0.85, PRPD_cross-correlation coefficient=0.75, ...] Trend feature extraction: System calculations revealed that although the absolute value of acetylene content was not high in the past 7 days, the growth rate was as high as 120%, showing an accelerating upward trend. At the same time, the statistical features of the PRPD spectrum (such as skewness and cross-correlation coefficient) have shifted significantly compared with a week ago.

[0052] S4: Intelligent Analysis and Early Warning Trigger The fused high-dimensional feature vectors are input into the pre-trained XGBoost multi-classification model and LSTM trend prediction model.

[0053] Output after comprehensive model analysis: 1. Failure probability: The probability of discharge failure is 82% (high risk threshold >75%).

[0054] 2. Health score: The equipment health score dropped from 92 points a week ago to 68 points.

[0055] 3. Trend prediction: The LSTM model predicts that, based on this trend, the acetylene content will exceed the warning value after 48 hours.

[0056] Since the probability of failure exceeds the high-risk threshold, the system immediately and automatically triggers a "critical" level warning.

[0057] S5: Event-based analysis and decision support Once the alert is triggered, the event analysis engine will automatically start: 1. Information aggregation: Automatically retrieve the equipment's register information: model, commissioning date, and factory test report.

[0058] It automatically links to and pops up a real-time video feed of the site, allowing maintenance personnel to remotely conduct a preliminary inspection of the equipment's appearance.

[0059] The transformer is automatically highlighted in the 3D model, and its air gap winding structure can be viewed with one click.

[0060] The system automatically searched the case database and found that a transformer of the same model from the same manufacturer had a similar discharge three years ago due to a manufacturing defect, which eventually developed into an inter-turn short circuit.

[0061] 2. Generate an event briefing: The system automatically generates a structured event briefing. Event Name: Extreme High-End YY Converter Acetylene Growth with Associated Partial Release Anomaly Incident severity: Severe Key evidence: Acetylene: 0.8 μL / L, a 120% increase over 7 days.

[0062] The PRPD spectrum shows typical suspended discharge characteristics.

[0063] Equipment of the same model has a history of family defects.

[0064] Preliminary assessment suggests the presence of floating potential discharge within the component, potentially related to loose connections or poor contact.

[0065] Recommendations for handling: Immediately shorten the oil chromatography tracking cycle to once every 4 hours.

[0066] It is recommended that the neutral point bushing connection and the iron core grounding lead be the focus of the next scheduled maintenance.

[0067] Strengthen infrared temperature monitoring and pay attention to localized overheating of the fuel tank.

[0068] S6: Visualization and Closed-Loop Processing The warning event and complete briefing information are pushed to the operation and maintenance personnel of the provincial monitoring center in a prominent manner through the "Smart Warning" card on the configurable large screen.

[0069] When maintenance personnel click on the event, the following will be displayed on the right side of the interface: Acetylene content trend curve (clearly showing the accelerating upward trend).

[0070] The slider button shows a comparison of the current PRPD graph with the graph from one week ago.

[0071] A schematic diagram of the location and internal structure of the equipment in the 3D model.

[0072] List of similar historical cases.

[0073] Based on the decision support provided by the system, maintenance personnel adopt the proposed solutions and issue instructions: 1. Adjust the oil chromatography monitoring frequency.

[0074] 2. Notify the on-site inspection personnel to conduct a special inspection of the equipment, paying particular attention to abnormal noises and infrared temperature measurement.

[0075] 3. Include this equipment in the key inspection items for the next power outage maintenance.

[0076] A week later, the transformer underwent scheduled maintenance. Based on precise analysis information from the system, maintenance personnel directly located and inspected the neutral point bushing connection, finding a slightly loose bolt. After tightening the bolt, the transformer was put back into operation, and since then, the acetylene content has stabilized, and the partial discharge signal has disappeared.

[0077] S7: Summary of Results This embodiment demonstrates the significant advantages of the present invention compared to traditional methods: 1. Early detection: Potential faults were detected several days in advance through multi-source fusion and trend analysis, even when the absolute value of acetylene was not exceeded and there were no traditional alarms.

[0078] 2. Accuracy: Through graph recognition and case database reasoning, the fault is accurately located to "floating discharge" and "loose connectors", which greatly shortens the on-site troubleshooting time.

[0079] 3. Intelligent: It automatically completes the entire process from data to early warning to handling suggestions, providing strong decision support for operation and maintenance personnel and realizing the transformation from "passive alarm" to "proactive early warning".

[0080] 4. Closed-loop system: Early warning information drives on-site inspections and maintenance, and the maintenance results are fed back to the system optimization model, forming a continuously improving intelligent operation and maintenance closed loop.

[0081] As can be seen from the specific embodiments, the method provided by the present invention has the following characteristics: early detection, discovering potential faults several days in advance through multi-source fusion and trend analysis when the absolute value of acetylene does not exceed the standard and there are no traditional alarms; accuracy, accurately locating faults to "floating discharge" and "loose connectors" through spectrum recognition and case library reasoning, greatly shortening the on-site investigation time; intelligence, automatically completing the entire process from data to early warning to handling suggestions, providing strong decision support for operation and maintenance personnel, realizing the transformation from "passive alarm" to "proactive early warning"; and closed-loop operation, with early warning information driving on-site inspection and maintenance, and the maintenance results being fed back to the system optimization model, forming a continuously improving intelligent operation and maintenance closed loop.

[0082] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0083] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0087] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0088] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0089] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0090] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A transformer fault early warning system based on multi-source heterogeneous data fusion analysis, characterized in that, The system includes: The data acquisition module is equipped with various data interfaces and adapters for real-time or near-real-time data acquisition from various heterogeneous data sources. The data platform module is used to clean, integrate, standardize, model, and manage the collected data. The intelligent analysis module is used to acquire data from the data platform module, perform in-depth mining, feature extraction, status assessment and fault prediction, and output early warning signals and judgment conclusions. The application service module is used to encapsulate various business capabilities in the form of microservices, including early warning services, reporting services, 3D services, and video services. Furthermore, these services are configured to be flexibly invoked and combined. The interactive module uses a micro-frontend architecture to build a visual interface, providing interactive entry points for various functions.

2. The transformer fault early warning system based on multi-source heterogeneous data fusion analysis according to claim 1, characterized in that, The data platform module includes a data lake and a data warehouse. The data lake is used to store raw data, and the data warehouse is used to store cleaned and processed standard data.

3. The transformer fault early warning system based on multi-source heterogeneous data fusion analysis according to claim 1, characterized in that, The intelligent analysis module incorporates various machine learning and data analysis algorithm models, including unsupervised learning models and supervised learning models; among them, Unsupervised learning models include clustering algorithms, which are used to cluster normal states and historical anomalous states to discover new anomalous patterns or to detect anomalies. Supervised learning models include classification models, regression models, and deep learning models. Classification models are used to train a multi-classification model to determine the type of failure; regression models are used to predict the remaining life of equipment or the future values ​​of key indicators; and deep learning models are used to perform deeper feature extraction and pattern recognition on time-series and image data using Long Short-Term Memory Networks (LSTM) and Convolutional Neural Networks (CNN).

4. A transformer fault early warning method based on multi-source heterogeneous data fusion analysis, characterized in that, The method uses the system described in any one of claims 1-3 to perform transformer fault early warning based on multi-source heterogeneous data fusion analysis.

5. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 4, characterized in that, The method includes: Collect multi-source heterogeneous data to comprehensively acquire all data related to transformer status; during collection, for real-time data streams, use message queues for subscription and push; for business data in relational databases, use data synchronization tools or scheduled tasks for incremental extraction; for file data, use file transfer services or object storage interfaces for uploading and retrieval; for third-party systems, obtain data by calling their provided RESTful API interfaces. The collected multi-source heterogeneous data is cleaned and standardized preprocessed. The cleaned multi-source data is deeply integrated in the data platform, and high-value features for fault early warning are extracted. The data fusion is carried out by combining feature-level fusion and decision-level fusion, and time-domain feature extraction, frequency-domain feature extraction, spectral feature extraction and trend feature extraction are performed according to feature engineering. By leveraging the fused high-dimensional features, intelligent analysis is performed through a pre-trained machine learning model to assess the status and provide early warnings of faults, outputting a health score or fault probability for the device. The warning results are transformed into actionable operation and maintenance suggestions and presented to users through the interface.

6. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 5, characterized in that, The collection of multi-source heterogeneous data to comprehensively acquire all data related to the transformer's condition includes: SCADA system is used to collect real-time operating data of transformers, circuit breakers and instrument transformers; The content and growth rate of each characteristic gas were collected using an oil chromatography monitoring system (DGA). A partial discharge monitoring system was used to collect UHF and ultrasonic AE signals to generate phase-resolved partial discharge PRPS, phase-resolved partial discharge pulse PRPD, and φ-QN spectra. The vibration acceleration and noise spectrum of the transformer tank are collected by a vibration acoustic monitoring system to analyze mechanical defects such as loose core clamps and winding deformation. The grounding current values ​​of the transformer core and clamps are collected by monitoring the grounding current of the core to determine whether there is a multi-point grounding fault. The pressure, density, and trace moisture content of the gas chamber in GIS equipment are monitored using an SF6 gas monitoring system. The auxiliary control system collects data from fire protection, security, environment, and video surveillance for comprehensive environmental risk assessment and alarm linkage. Real-time meteorological data of the site area is obtained using a meteorological system to analyze the impact of the external environment on the thermal stability and insulation performance of the equipment. Static data such as equipment ledger information, test reports, maintenance history, and family defect records are obtained through the Production Management System (PMS) to provide historical basis for condition assessment. When a fault occurs, the fault recording system collects and parses the fault recording file to obtain the waveform, amplitude, phase, and protection action sequence of the fault current and voltage. The robot inspection system receives infrared thermal images and visible light images transmitted back by the inspection robot to identify defects such as overheating and abnormal appearance of the equipment.

7. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 5, characterized in that, The cleaning and standardization preprocessing of the collected multi-source heterogeneous data includes: For outliers, business rules and statistical methods are used to identify and handle abnormal fluctuations and singularities; For null values, methods such as deletion, filling with the mean / median, or filling based on machine learning algorithm prediction are used, depending on the importance of the data. Standardize the format, unify the timestamp format, numerical units, and enumeration value encoding; Remove duplicate data caused by network retransmission; For unstructured data, feature extraction is performed to convert it into structured labels; Time alignment is performed for data streams of different frequencies; Additionally, data with different dimensions are scaled to the same numerical range to eliminate the influence of dimensions between features.

8. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 5, characterized in that, The feature-level fusion includes concatenating low-level feature vectors from different data sources into a high-dimensional joint feature vector; the decision-level fusion includes first performing independent analysis and preliminary decision-making on different data sources or models, and then fusing the decision results; the time-domain feature extraction includes calculating statistical features such as mean, variance, peak value, and kurtosis; the frequency-domain feature extraction includes performing Fast Fourier Transform (FFT) on vibration and acoustic signals to extract the fundamental frequency, harmonic components, and frequency centroid; the spectral feature extraction includes extracting skewness, steepness, cross-correlation coefficient, and discharge phase distribution from the partial discharge PRPD spectrum; and the trend feature extraction includes calculating the short-term and long-term rates of change of key indicators and the sliding window average value to capture the deterioration trend of the indicators.

9. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 5, characterized in that, The assessment status and early warning faults include: setting multi-level early warning thresholds, and automatically triggering an early warning of the corresponding level when the score exceeds the threshold. The early warning is set to take into account both instantaneous value exceeding the limit and trend warning.

10. The transformer fault early warning method based on multi-source heterogeneous data fusion analysis according to claim 5, characterized in that, The content presented to the user through the interface includes: It adopts micro-frontend and configurable large screen technology, allowing users to customize the monitoring interface by dragging and dropping components; It offers a rich set of visualization components, including David's triangle and three-ratio plots for visualizing oil chromatography data analysis results, PRPD / PRPS spectrum display components for interactive analysis of partial discharge types, trend comparison curves that support trend comparison analysis of multiple parameters and time periods, and 3D model linkage with decision support wizards.

Images