Zone area health detection system and method based on edge calculation

Abstract

The invention discloses an area health detection system and method based on edge calculation, and relates to the technical field of intelligent monitoring of power systems. According to the method, the edge computing terminal is deployed in the power distribution area, real-time monitoring and health assessment of the operation state of the area are achieved, and support is provided for operation and maintenance decision making in combination with big data analysis and model prediction of a master station platform. The system has the advantages of being real-time, efficient, comprehensive in perception, intelligent in decision making, edge autonomy, edge cloud collaboration and the like, and the operation management level and the power supply reliability of a power distribution area can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power systems, specifically to a distribution transformer health monitoring system and method based on edge computing, which is applied to real-time health status monitoring of distribution transformer areas, identification of potential operational hazards, guidance of operation and maintenance decisions, and improvement of power supply reliability. Background Technology

[0002] The statements in this section are provided only as background information in relation to this disclosure and may not constitute prior art.

[0003] Distribution transformer substations are the final link in the distribution network, directly supplying power to users. Their operational status is crucial to power quality and reliability. Traditional substation health monitoring mainly relies on manual inspections and periodic testing, which suffers from low efficiency and poor real-time performance, making it difficult to accurately reflect the real-time operating status of equipment.

[0004] In recent years, with the development of the Internet of Things (IoT) and sensor technology, some power distribution areas have begun to deploy online monitoring devices. However, existing systems often lack edge computing capabilities, have limited data acquisition and processing capabilities, and cannot comprehensively and efficiently acquire the electrical characteristic parameters of the distribution areas. Real-time analysis is insufficient, making it difficult to meet the power system's requirements for rapid response. Furthermore, their reliability is low; once the network connection is interrupted, the cloud processing mode completely fails, and the detection capability is lost. The data value density is low; the uploaded raw data contains a large amount of useless information, wasting storage and computing resources. Therefore, a new detection system and method are needed that can achieve local intelligent processing, real-time response, and collaboration with the cloud.

[0005] Furthermore, traditional evaluation systems lack unified standards, with indicator selection and weight allocation relying heavily on expert experience, resulting in strong subjectivity and insufficient scientific rigor. Simultaneously, manual evaluation methods suffer from low levels of intelligence, leading to inefficiency and susceptibility to human error. These issues make it difficult for traditional methods to achieve real-time, objective assessments of transformer area health status, thus failing to provide effective data support for operational and maintenance decisions. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in the prior art by providing a transformer substation health monitoring system and method based on edge computing. This system, through an edge-cloud collaborative architecture, pushes simple computations to the edge terminal, while analysis is handled by the main station platform. This enables real-time monitoring and health assessment of the operating status of distribution transformer substations, improving operational efficiency and power supply reliability, and enhancing the real-time performance, accuracy, and intelligence of transformer substation health monitoring.

[0007] The technical solution of the present invention is as follows: A transformer area health monitoring system based on edge computing includes: An edge computing terminal device is deployed at the distribution substation site and has a built-in edge computing module for collecting and processing substation operation data on-site. The operation data includes electrical parameters, equipment status, sensor data, environmental parameters, and image data. The edge computing module performs real-time analysis on the operation data and extracts features to output the substation health index and sub-health index. The main platform, deployed in the cloud or on a central server, is used to receive the processing results and key data uploaded by the edge computing terminal devices, run machine learning models and expert systems to predict the health trend of the transformer area, assess the risk of failure and generate operation and maintenance suggestions; it can also distribute the analysis results to the edge computing terminal devices to achieve edge-cloud collaboration. The communication network connects the edge computing terminal device and the main station platform, supports bidirectional data transmission, and the communication protocols include DLMS, IDIS, MQTT, and HTTP. The communication methods include Ethernet, 4G / 5G wireless public network, and optical fiber.

[0008] Furthermore, the edge computing terminal device includes: Multi-core microprocessor, memory, storage medium, communication module, electrical data sampling module, input / output module, and power supply module; Among them, the multi-core microprocessor is the core of the edge computing terminal device, responsible for running the operating system and applications, and performing data processing and model inference tasks; The memory is used to store temporary operating data and a health assessment model, which is used to calculate the health index of the transformer area in real time based on the extracted features. The storage medium is used to permanently store system firmware, configuration parameters, and historical data. The communication module includes: Ethernet, 4G / 5G uplink module, GPS / BeiDou, high-speed power line carrier, and radio frequency wireless, used for communication with the main station platform and other devices; The electrical data sampling module includes: a CT current sampling module, a PT voltage sampling module, and a metering chip; The input / output module includes: an extended CAN bus interface module for acquiring sensor data, a serial port, an RS485 interface for connecting to smart meters or sensors, a digital input / output interface for acquiring switch status and controlling indicator lights or alarms, and an infrared thermal imager. The power module converts external power into a stable power supply required by various terminal components, and is equipped with a backup battery and supercapacitor to ensure that it can still operate for a short time and report critical information when external power is lost.

[0009] Furthermore, the edge computing terminal device processes the collected station operation data as follows: Step A: Preprocess the collected transformer area operation data; Step B: Extract features from the preprocessed data to generate key features that reflect the operating status of the transformer area; Step C: The extracted multidimensional feature vectors are input into the health assessment model for inference.

[0010] Furthermore, the preprocessing includes: Data cleaning involves removing obviously abnormal or erroneous sampled values. Time synchronization ensures that data from all sensors are aligned in time. Standardization converts parameters with different dimensions into a unified dimensionless value; Normalization scales numerical values ​​to the range required by the model input. The preprocessed data is stored in memory for subsequent feature extraction and model inference.

[0011] Furthermore, during the feature extraction process, appropriate feature extraction methods are adopted for different types of data, as detailed below: Sliding window statistics and sequence rate of change are used to analyze the time-series characteristics of electrical parameters; One-dimensional convolutional neural network (1DCNN) is used to extract frequency domain features for harmonic characteristics; The trend slope and temperature rise rate are calculated based on the oil temperature / oil level trend characteristics. Extracting statistical features from environmental parameter characteristics; Temperature distribution features are extracted from image features.

[0012] Furthermore, the health assessment model employs a simplified multilayer perceptron model, which has already learned the mapping relationship between features and health indices during the training phase. The multilayer perceptron model outputs the current health index (HI) of the station area and the health scores of each subsystem.

[0013] Furthermore, the Health Index (HI) is calculated using a weighted composite method: HI=100-( ×Voltage imbalance rate+ ×Harmonic distortion rate+ ×Number of times oil temperature exceeded limit+ ×Number of low fuel level alarms+ ×Environmental anomaly rate… ) in: arrive These are the weighting coefficients for each indicator. This value is determined by training a meta-model, which can be optimized based on expert experience and historical data. Edge computing terminal devices calculate health indices in real time and classify health status into different levels based on preset thresholds.

[0014] Furthermore, the edge computing terminal device performs anomaly detection based on the health assessment results; When the health index is lower than a certain threshold or the score of a certain subsystem is lower than the corresponding threshold, it is determined that there is an abnormality or potential fault risk in the area, and the terminal immediately triggers the local alarm mechanism. The local alarm mechanism includes: on the one hand, emitting sound and light signals through the built-in sound and light alarm to alert on-site personnel; on the other hand, sending alarm information to the pre-set mobile phones of maintenance personnel or monitoring platforms via SMS or network communication, including the station area number, alarm type, time of occurrence, and current health index. Alarm events are also recorded in local logs and databases to save key data and characteristics at the time, so as to facilitate subsequent analysis of the cause of the failure and optimization of model parameters; Edge computing terminal devices upload data to the main platform according to a set period when important events occur; the uploaded data includes: fragments of raw sampled data, extracted feature vectors, calculated health index time series, and alarm records; After receiving the data, the main station platform first stores it in the database for long-term preservation, forming a historical archive of the station's operation. Subsequently, the main station platform's data processing module further cleans and integrates the data, and performs feature engineering to generate input suitable for model training.

[0015] Furthermore, the main platform deploys various machine learning models and analysis tools to achieve in-depth integrated analysis of the health status of the transformer substations (see [link]). Figure 5 (Main site platform risk analysis process) Health trend prediction: The model uses a long short-term memory network or a Transformer model to train historical health index sequences and learn their temporal evolution patterns. The trained model can predict the health index trend in the next few hours to days and generate a trend prediction curve. Fault Risk Assessment: The XGBoost ensemble learning model is used, combined with historical fault case data, to train the multi-source features of the transformer area and output the probability or risk level of a fault occurring in the future. The XGBoost model can handle high-dimensional features and nonlinear relationships, and accurately assess the impact of various factors on fault risk. The main station platform can set different risk levels and automatically determine the risk level of the transformer area based on the model output for reference in operation and maintenance decisions. Periodic anomaly identification: Using time series analysis and pattern recognition techniques, detect whether there are periodic anomalies in the operation data of the transformer area; Operation and maintenance decision support: The main platform integrates the above analysis results and generates operation and maintenance suggestions. These suggestions are provided to operation and maintenance managers through reports or visualization interfaces as a basis for formulating maintenance plans and optimizing operation strategies.

[0016] This invention also proposes a method for detecting the health of transformer substations based on edge computing, comprising the following steps: Edge computing terminal devices are deployed on-site in the distribution area to collect and preprocess multi-source operational data in real time; Features are extracted from the edge side and a health index is calculated using an MLP model; A local alarm is triggered when health indicators are abnormal. Upload key data and health assessment results to the main platform; The main platform uses machine learning models to predict health trends, assess fault risks, and identify periodic anomalies, and generates operation and maintenance suggestions. The main platform distributes the analysis results to edge computing terminal devices to achieve edge-cloud collaboration (see cloud-edge collaboration timing sequence). Figure 6 ).

[0017] Compared with existing technologies, the advantages of this invention are: 1. Real-time performance and high efficiency: By deploying edge computing terminals on-site in the transformer substations, data collection and processing are achieved locally, which greatly reduces data transmission latency and cloud computing pressure. It can calculate the health index of the transformer substations in real time and issue abnormal alarms in a timely manner, meeting the power system's requirements for rapid response.

[0018] 2. Comprehensive Perception and Precise Assessment: By integrating multi-source data such as electrical parameters, equipment status, environmental parameters, and images, a comprehensive health evaluation index system for transformer substations has been constructed, which can fully reflect the operational status of the substations. The use of a combination of machine learning models and feature engineering improves the objectivity and accuracy of health assessments, avoiding the subjective biases of traditional expert judgment.

[0019] 3. Intelligent Decision Support: The main station platform utilizes advanced machine learning algorithms to analyze long-term data, predicting health trends, identifying potential failure risks, and providing operational and maintenance suggestions, thus offering a scientific basis for operation and maintenance and decision-making. The edge-cloud collaborative architecture combines local monitoring with remote analysis, improving the intelligence level of transformer area operation and maintenance.

[0020] 4. Reduced Operation and Maintenance Costs: This invention enables automatic monitoring and assessment of the health status of transformer substations, reducing the workload and frequency of manual inspections and lowering operation and maintenance costs. Simultaneously, by utilizing advanced AI models on the main station platform, trend prediction and risk warning are performed, achieving a shift from "post-fault repair" to "pre-fault warning," thus preventing power outages caused by escalating faults and improving power supply reliability and economic efficiency.

[0021] 5. Comprehensive Improvement in Reliability: Real-time processing at the edge reduces reliance on the network, ensuring core monitoring capabilities even during network outages. Post-fault repair becomes pre-fault early warning, significantly enhancing the reliability of power supply.

[0022] 6. Edge-cloud collaboration and complementary advantages: The edge side handles real-time, simple tasks, while the main site handles long-term, complex analyses, achieving optimized allocation of computing resources while ensuring depth of analysis. This maximizes returns at low cost. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall architecture of the edge computing-based health monitoring system for transformer substations as described in this invention.

[0024] Figure 2 This is a schematic diagram of the process for calculating the health index of the edge terminal in the distribution area as described in this invention.

[0025] Figure 3 This is a schematic diagram of the hardware structure of the edge computing terminal device described in this invention.

[0026] Figure 4 This is a schematic diagram of the software functional modules of the main station platform described in this invention.

[0027] Figure 5 This is a schematic diagram of the risk analysis process of the main site platform described in this invention.

[0028] Figure 6 This is a schematic diagram of the collaborative working timing described in this invention. Detailed Implementation

[0029] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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 a process, method, article, or apparatus. Without further limitations, 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 said element.

[0030] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0031] Example 1 Please see Figures 1-6 A health monitoring system for transformer substations based on edge computing, comprising: Edge computing terminal devices are deployed in power distribution areas (e.g., installed in distribution boxes or cabinets near power distribution transformers). They have built-in edge computing modules for collecting and processing area operation data locally. This operation data includes multi-source sensing data such as electrical parameters, equipment status, sensor data, environmental parameters, and image data. The edge computing module performs real-time analysis and feature extraction on the operation data, outputting an area health index and sub-health indices. In other words, the terminal device has built-in high-performance embedded processors, memory, and communication modules, possessing edge computing capabilities, enabling local data collection, processing, and preliminary analysis, achieving real-time feature extraction and health scoring. The main platform, deployed in the cloud or on a central server (e.g., in a cloud data center or a power company's monitoring center), consists of a server cluster and a database. It receives processing results and key data uploaded by the edge computing terminal devices, runs machine learning models (such as LSTM, Transformer, XGBoost, etc.) and expert systems to predict the health trend of the distribution area, assess fault risks, and generate operation and maintenance suggestions. It can also distribute analysis results (such as trend predictions and maintenance suggestions) to the edge computing terminal devices to achieve edge-cloud collaboration. The communication network connects the edge computing terminal device and the main station platform, supports bidirectional data transmission, and the communication protocols include DLMS, IDIS, MQTT, HTTP, etc. The communication methods include Ethernet, 4G / 5G wireless public network, fiber optic, etc.; that is, the communication network adopts a secure and reliable transmission protocol (such as DLMS, IDIS, MQTT, HTTP, etc.) to support bidirectional communication between the edge terminal and the main station, ensuring the real-time performance and security of data transmission.

[0032] In this embodiment, it should be noted that, please refer to... Figure 3 The edge computing terminal device includes: Multi-core microprocessor (MPU), memory (RAM), flash storage medium, communication module, electrical data sampling module, input / output module, and power supply module; Among them, the multi-core microprocessor is the core of the edge computing terminal device, responsible for running the operating system and applications, and performing data processing and model inference tasks; The memory is used to store temporary operating data and a health assessment model, which is used to calculate the health index of the transformer area in real time based on the extracted features. The storage medium is used to permanently store system firmware, configuration parameters, and historical data. The communication module includes: Ethernet, 4G / 5G uplink module, GPS / BeiDou, high-speed power line carrier (HPLC), and radio frequency (RF) for communicating with the main station platform and other devices; The electrical data sampling module includes: a CT current sampling module, a PT voltage sampling module, and a metering chip; Input / output modules include: an extended CAN bus interface module for acquiring sensor data, a serial port, an RS485 interface for connecting smart meters or sensors, a digital input / output interface for acquiring switch status and controlling indicator lights or alarms, an infrared thermal imager, etc. The power module converts external power into a stable power supply required by various terminal components, and is equipped with a backup battery and supercapacitor to ensure that it can still operate for a short time and report critical information when the external power is interrupted. In this embodiment, it should be noted that in the distribution area, the edge computing terminal device collects multi-source data in real time through various sensors and interfaces; For example, through the high-precision CT current sampling module, PT voltage sampling module and metering chip of the electrical data sampling module, electrical parameters such as three-phase voltage, current, active power, reactive power and power factor of the low-voltage side of the transformer can be collected. The open / closed status of the main switch and each branch switch is obtained through the auxiliary contacts of the switch or the intelligent circuit breaker; Transformer oil temperature and oil level data are obtained through temperature sensors and oil level sensors; The temperature, humidity, and alarm signals such as smoke and water immersion in the power distribution room are obtained through environmental sensors. The transformer and key connection points are periodically photographed using an infrared thermal imager to detect abnormal heating. All collected data enters the system through the input interface and is processed within the terminal.

[0033] In this embodiment, specifically, the edge computing terminal device processes the collected station operation data as follows: Step A: Preprocess the collected transformer area operation data; Step B: Extract features from the preprocessed data to generate key features that reflect the operating status of the transformer area; Step C: The extracted multidimensional feature vectors are input into the health assessment model for inference.

[0034] In this embodiment, the preprocessing steps specifically include: data cleaning (removing obviously abnormal or erroneous sampled values), time synchronization (ensuring that the data from each sensor are aligned in time), standardization (converting parameters with different dimensions into uniform dimensionless values), and normalization (scaling the values ​​to the range required by the model input, such as the 0-1 interval). The preprocessed data is stored in the terminal's memory for subsequent feature extraction and model inference.

[0035] The edge computing terminal extracts features from the preprocessed data to generate key features that reflect the operational status of the distribution area. Appropriate feature extraction methods are used for different types of data, as detailed below: Sliding window statistics and sequence rate of change are used to analyze the time-series characteristics of electrical parameters; One-dimensional convolutional neural network (1DCNN) is used to extract frequency domain features for harmonic characteristics; The trend slope and temperature rise rate are calculated based on the oil temperature / oil level trend characteristics. Extracting statistical features from environmental parameter characteristics; Temperature distribution features are extracted from image features.

[0036] Please see Figure 2 The details are as follows: Electrical parameter time-series characteristics: For electrical parameters such as voltage, current, and power that vary over time, a sliding window technique is used to extract a time series of a certain duration (e.g., 5 minutes). Statistical characteristics within the window are calculated, such as average, maximum, minimum, root mean square (RMS), and volatility (standard deviation). Simultaneously, the rate of change between adjacent windows is calculated to capture the trend of parameter changes. These characteristics can reflect the load level of the transformer substation, the three-phase balance, and voltage stability.

[0037] Harmonic Features: For harmonic content data, the content rate of each harmonic (e.g., 5th, 7th, 11th, etc.) is grouped into a vector to form spectral features. To extract frequency domain patterns using deep learning, the harmonic vector can be converted into a one-dimensional "spectral graph" and input into a one-dimensional convolutional neural network (1DCNN) for feature extraction. 1DCNN can automatically learn the patterns in the harmonic spectrum and output a quantitative representation of the degree of harmonic distortion.

[0038] Oil temperature / level trend characteristics: Transformer oil temperature, oil level, and other parameters change relatively slowly; the main focus is on their long-term trends and abnormal fluctuations. The slope of oil temperature change (temperature rise per unit time) and the rate of oil level decrease are calculated as derived characteristics. If the oil temperature rises abnormally within a short period or the oil level continues to drop, these characteristic values ​​will change significantly, indicating a potential fault.

[0039] Environmental parameter characteristics: Statistical features such as mean, maximum / minimum values, and rate of change over a period of time are extracted for environmental parameters like temperature and humidity. For example, excessively high temperatures or humidity in a power distribution room may affect the insulation performance of equipment; these characteristics can be used to assess the impact of the environment on equipment health.

[0040] Image features: Image processing algorithms are used to extract temperature distribution characteristics of key areas from infrared thermal imaging images. For example, the maximum temperature values ​​of transformer windings, bushings, and terminals can be identified and compared with historical images to detect any abnormal hot spots. Image features can be used as auxiliary information input into health assessment models to improve the ability to detect thermal faults in equipment.

[0041] In this embodiment, specifically, the extracted multidimensional feature vector is input into the health assessment model for inference; The health assessment model employs a simplified Multilayer Perceptron (MLP) model, which has already learned the mapping relationship between features and health indices during the training phase. The MLP model outputs the current health index HI (e.g., 0-100 points) for the distribution area, as well as the health scores for each subsystem (e.g., electrical performance score, equipment status score, environmental adaptability score, etc.). The health index can also be calculated using a weighted aggregation method, for example: HI=100-( ×Voltage imbalance rate+ ×Harmonic distortion rate+ ×Number of times oil temperature exceeded limit+ ×Number of low fuel level alarms+ ×Environmental anomaly rate… ) in: arrive This refers to the weighting coefficients of each indicator, which are determined by training a meta-model and can be optimized based on expert experience and historical data. The formula sums the weighted deterioration indicators according to their weights and subtracts them from 100 points to obtain the final health index. The lower the score, the worse the health status of the transformer area. The edge computing terminal calculates the health index in real time and classifies the health status into different levels (such as "healthy", "attention", "alarm" etc.) according to preset thresholds. In this embodiment, it should also be noted that the specific method for determining the weights through meta-model learning is as follows: Using a shallow neural network, the outputs of each sub-model … Using historical fault data or expert-annotated overall status as input, the system is trained to automatically learn the optimal weight allocation. ... Then, experts fine-tune and determine the weights. The same learning method can also be used to determine the weights in risk assessment; in this embodiment, Indicates the voltage imbalance rate. Indicates harmonic distortion rate. This indicates the number of times the oil temperature exceeded the limit. This indicates the number of low oil level alarms. This indicates the rate of environmental anomalies.

[0042] In this embodiment, specifically, the edge computing terminal device performs anomaly detection based on the health assessment results; When the health index is below a certain threshold (e.g., below 70 points) or the score of a certain subsystem is below the corresponding threshold, it is determined that there is an abnormality or potential fault risk in the station area, and the terminal immediately triggers the local alarm mechanism. The local alarm mechanism includes: on the one hand, emitting sound and light signals through the built-in sound and light alarm to remind on-site personnel to pay attention; on the other hand, sending alarm information to the pre-set mobile phones of maintenance personnel or monitoring platforms via SMS or network communication, including the transformer area number, alarm type (such as "oil temperature too high", "voltage abnormal", etc.), occurrence time, and current health index. In addition, the terminal also records alarm events to local logs and databases, saving key data and characteristics at the time, so as to analyze the cause of the failure and optimize model parameters later; local alarms ensure that abnormal situations can be detected and responded to in the first time, preventing small problems from developing into major failures.

[0043] In this embodiment, specifically, the edge computing terminal device uploads data to the main station platform according to a set period when an important event occurs; the uploaded data includes: fragments of raw sampling data (such as voltage and current waveform data of the most recent hour), extracted feature vectors, calculated health index time series, alarm records, etc. After receiving the data, the main station platform first stores it in the database for long-term preservation, forming a historical archive of the station's operation. Subsequently, the main station platform's data processing module further cleans and integrates the data, and performs feature engineering to generate input suitable for model training.

[0044] In this embodiment, specifically, the main station platform deploys various machine learning models and analysis tools to achieve in-depth fusion analysis of the health status of the transformer substations (see...). Figure 5 ): Health trend prediction: The model is trained on historical health index sequences using Long Short-Term Memory (LSTM) or Transformer models to learn their temporal evolution patterns. The trained model can predict the health index trend in the next few hours to days and generate a trend prediction curve. For example, it can predict whether the health index will continue to decline in the next 7 days, thus providing an early warning of possible health deterioration. Fault Risk Assessment: Employing ensemble learning models such as XGBoost, combined with historical fault case data, the system trains on the multi-source features of the distribution area to output the probability or risk level of a fault occurring in the future. The XGBoost model can handle high-dimensional features and nonlinear relationships, accurately assessing the impact of various factors on fault risk. The main station platform can set different risk levels (such as low risk, medium risk, and high risk) and automatically determine the risk level of the distribution area based on the model output for reference in operation and maintenance decisions. Periodic anomaly identification: Using time series analysis and pattern recognition techniques, the system detects whether periodic anomalies exist in the operating data of the transformer substation. For example, through Fourier transform or autocorrelation analysis, it identifies whether the harmonic content fluctuates at a fixed frequency every week, or whether the oil temperature is regularly higher in the afternoon each day. Once a periodic anomaly pattern is detected, the main station platform will record it and alert the maintenance personnel so that targeted measures can be taken (such as adjusting the operating mode or arranging maintenance). Operation and Maintenance Decision Support: The main platform integrates the above analysis results to generate operation and maintenance suggestions. For example, if the health index of a certain transformer area is predicted to continue to decline in the coming week and the fault risk assessment is high, it is recommended to arrange maintenance as soon as possible. If a periodic increase in harmonic distortion is found, which may be caused by the periodic load of an industrial user, it is recommended to monitor or manage the user's electricity usage behavior. The main platform can also provide suggestions for optimizing operation based on historical data and expert knowledge base, such as adjusting transformer taps to improve voltage quality and shifting some loads to reduce heavy-load phase current. These suggestions are provided to operation and maintenance managers through reports or visualization interfaces as a basis for formulating maintenance plans and optimizing operation strategies.

[0045] In this embodiment, it should also be noted that after the main station platform completes the analysis, it sends the key results to the edge computing terminal device; The distributed information includes: health trend prediction curves, risk levels for a future period, descriptions of identified abnormal patterns, and maintenance recommendations. After receiving this information, the edge computing terminal stores it locally and displays it to on-site maintenance personnel through a graphical interface. For example, the terminal can be equipped with a display screen showing the predicted trend of the health index for the next 24 hours, as well as current risk level prompts (e.g., "High risk, please check as soon as possible"). Simultaneously, the main platform also centrally displays the health status, alarm information, and analysis results of each transformer substation through the power company's integrated monitoring system for remote monitoring personnel to view. The monitoring interface can be a GIS map, marking the location and health status color of each substation (e.g., green for healthy, yellow for caution, red for alarm). Clicking on a specific substation allows viewing detailed health index curves, alarm records, and recommended measures. Through this edge-cloud collaborative mechanism, on-site and remote operations achieve information sharing and linkage: the edge side ensures real-time monitoring and rapid response, while the cloud provides a global perspective and in-depth analysis; the combination of both significantly improves the intelligence level of substation health management.

[0046] In this embodiment, it should also be noted that the main platform regularly uses newly accumulated data to train and update the machine learning model in order to ensure the accuracy and adaptability of the model. The training process includes data preparation, feature selection, model training, validation, and deployment. The main platform can use distributed computing frameworks (such as TensorFlow or PyTorch) to train complex models, such as deep neural networks, on a server cluster. After training, the model performance is evaluated, and if it meets the requirements, it is exported to an optimized format (such as ONNX or TensorFlow Lite) and distributed to edge computing terminal devices. Upon receiving the updated model, the edge terminal loads it locally and replaces the old model for inference. In this way, the main platform can continuously optimize the model algorithm, and the edge terminals can obtain the latest model capabilities in a timely manner, achieving continuous improvement in system performance. Simultaneously, the edge terminals running simplified models locally ensure that even if communication with the main platform is interrupted, they can still independently complete health checks and alarm functions, improving system reliability.

[0047] Furthermore, this invention fully considers security and reliability in its system design. Data transmission between the edge computing terminal device and the main station platform employs encryption protocols (such as TLS / SSL) and authentication mechanisms to prevent data theft or tampering. The terminal device has local storage and offline operation capabilities, allowing it to continue collecting and storing data even during communication interruptions. Once communication is restored, the data is retransmitted to the main station, ensuring data integrity. The terminal hardware features self-testing and redundancy designs; for example, critical sensors employ dual-path backup, and important data is periodically backed up to storage media to prevent data loss due to single-point failures. In addition, the system implements multi-level access control, ensuring that only authorized personnel can access and configure the terminal device and the main station platform, guaranteeing the security of system operation.

[0048] This invention also proposes a method for detecting the health of transformer substations based on edge computing, comprising the following steps: Data acquisition and preprocessing: Edge computing terminal devices are deployed on-site in the distribution area to collect and preprocess multi-source operational data in real time; Feature extraction and health assessment: Features are extracted at the edge and a health index is calculated using an MLP model; Anomaly detection and local alarms: Trigger local alarms when health indicators are abnormal; Data upload and cloud analysis: Upload key data and health assessment results to the main platform; Edge-cloud collaboration and results feedback: The main platform uses machine learning models to predict health trends, assess fault risks, and identify periodic anomalies, and generates operation and maintenance suggestions; The main platform distributes the analysis results to edge computing terminal devices to achieve edge-cloud collaboration.

[0049] In this embodiment, specifically in the data acquisition and preprocessing stage: edge computing terminal devices are deployed on-site in the distribution transformer area to collect real-time operational data of the area through sensors and interfaces. The collected data includes: electrical parameters such as transformer phase voltage, current, active / reactive power, and harmonic content; equipment status such as the opening and closing status of main and branch switches; sensor data such as transformer oil temperature and level; environmental parameters such as ambient temperature, humidity, and smoke; and image data such as infrared thermal imaging photos. The edge terminal preprocesses the collected raw data, including data cleaning, standardization, and normalization, to eliminate noise and dimensional influences, preparing for subsequent analysis.

[0050] In this embodiment, specifically in the feature extraction and health assessment stage: the edge computing terminal device performs feature extraction and health assessment on the preprocessed data. Depending on the data type, corresponding feature engineering methods are employed: for time-series data such as electrical parameters, sliding window statistics (e.g., mean, maximum, volatility) and sequence change rate methods are used to extract features to capture the trend and fluctuation characteristics of the data; for harmonic data, the content of each harmonic is converted into a spectrum diagram, and frequency domain features are extracted using a one-dimensional convolutional neural network (1DCNN); for low-frequency trend data such as oil temperature and oil level, derived features such as the slope of the change trend and the rate of temperature rise are calculated; for environmental parameters, statistical features (e.g., mean, extreme values, rate of change) are extracted. The extracted multidimensional features are input into a pre-trained simplified health assessment model (multilayer perceptron MLP model) for inference, obtaining the current health index (HI) of the transformer substation and sub-health indices for each dimension. The health index can be represented on a percentage scale (0-100 points), with higher scores indicating a healthier operating status for the transformer substation.

[0051] In this embodiment, specifically during the anomaly detection and local alarm phase: the edge computing terminal performs anomaly detection and local alarms based on the health assessment results. When the health index falls below a set threshold or a certain sub-health index becomes abnormal, the terminal device immediately triggers a local alarm, alerting maintenance personnel via an audible and visual alarm, and notifying relevant personnel via SMS, APP push notifications, etc. Simultaneously, the terminal device records the alarm event and timestamp, forming a historical alarm log for subsequent analysis.

[0052] In this embodiment, specifically, data uploading and cloud analysis occur as follows: The edge computing terminal uploads key data and health assessment results to the main station platform. The uploaded data includes raw sampled data fragments, extracted feature vectors, health indices, and alarm records. The main station platform stores and further analyzes the received data: it uses machine learning models (such as LSTM, Transformer, and other time-series models) to train historical health indices and predict health trends over a future period; it uses XGBoost models or expert systems to perform fusion analysis on the uploaded multi-source sensing data to assess the fault risk level of the transformer area within a specific future time window; simultaneously, the main station platform can also identify periodic abnormal patterns (such as harmonic exceedances or excessively high oil temperatures occurring at fixed times each week). The main station platform summarizes the analyzed trend prediction curves, risk levels, abnormal patterns, and other results to generate maintenance recommendations (such as adjusting load distribution and scheduling maintenance).

[0053] In this embodiment, specifically during the edge-cloud collaboration and result feedback phase: the main station platform sends the analysis results and operation and maintenance suggestions to the edge computing terminal devices. Upon receiving these, the edge terminals locally display and store information such as trend prediction curves and risk warnings for on-site operation and maintenance personnel to reference. Simultaneously, the main station platform can also display an overview of the health status, trend predictions, and alarm information for each transformer substation to the remote operation and maintenance center through a graphical interface, enabling remote monitoring and decision support for the substation's operational status. Through this edge-cloud collaboration mechanism, the edge side is responsible for real-time monitoring and rapid response, while the cloud side is responsible for big data analysis and model optimization. The two complement each other's strengths, jointly improving the level of substation health management.

[0054] In summary, the edge computing-based transformer substation health monitoring system and method provided by this invention achieves real-time monitoring and health assessment of the substation's operational status by deploying edge computing terminals in the substation. Combined with big data analysis and model prediction from the main station platform, it provides support for operation and maintenance decisions. This system offers advantages such as real-time efficiency, comprehensive perception, intelligent decision-making, high reliability, and edge-cloud collaboration, significantly improving the operation and management level and power supply reliability of the substation.

[0055] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

[0056] This background section is provided to generally present the context of the invention. The work of the currently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither expressly nor impliedly acknowledged as prior art to the invention.

Claims

1. A transformer area health monitoring system based on edge computing, characterized in that, include: An edge computing terminal device is deployed at the distribution substation site and has a built-in edge computing module for collecting and processing substation operation data on-site. The operation data includes electrical parameters, equipment status, sensor data, environmental parameters, and image data. The edge computing module performs real-time analysis on the operation data and extracts features to output the substation health index and sub-health index. The main platform, deployed in the cloud or on a central server, is used to receive the processing results and key data uploaded by the edge computing terminal devices, run machine learning models and expert systems to predict the health trend of the transformer area, assess the risk of failure, and generate operation and maintenance suggestions. The analysis results can also be sent to edge computing terminal devices to achieve edge-cloud collaboration; The communication network connects the edge computing terminal device and the main station platform, supports bidirectional data transmission, and the communication protocols include DLMS, IDIS, MQTT, and HTTP. The communication methods include Ethernet, 4G / 5G wireless public network, and optical fiber.

2. The edge computing-based health monitoring system for transformer substations according to claim 1, characterized in that, The edge computing terminal device includes: Multi-core microprocessor, memory, storage medium, communication module, electrical data sampling module, input / output module, and power supply module; Among them, the multi-core microprocessor is the core of the edge computing terminal device, responsible for running the operating system and applications, and performing data processing and model inference tasks; The memory is used to store temporary operating data and a health assessment model, which is used to calculate the health index of the transformer area in real time based on the extracted features. The storage medium is used to permanently store system firmware, configuration parameters, and historical data. The communication module includes: Ethernet, 4G / 5G uplink module, GPS / BeiDou, high-speed power line carrier, and radio frequency wireless, used for communication with the main station platform and other devices; The electrical data sampling module includes: a CT current sampling module, a PT voltage sampling module, and a metering chip; The input / output module includes: an extended CAN bus interface module for acquiring sensor data, a serial port, an RS485 interface for connecting to smart meters or sensors, a digital input / output interface for acquiring switch status and controlling indicator lights or alarms, and an infrared thermal imager. The power module converts external power into a stable power supply required by various terminal components, and is equipped with a backup battery and supercapacitor to ensure that it can still operate for a short time and report critical information when external power is lost.

3. The edge computing-based health monitoring system for transformer substations according to claim 2, characterized in that, The edge computing terminal device processes the collected station operation data as follows: Step A: Preprocess the collected station operation data; Step B: Extract features from the preprocessed data to generate key features that reflect the operating status of the transformer area; Step C: The extracted multidimensional feature vectors are input into the health assessment model for inference.

4. The edge computing-based health monitoring system for transformer substations according to claim 3, characterized in that, The preprocessing includes: Data cleaning involves removing obviously abnormal or erroneous sampled values. Time synchronization ensures that data from all sensors are aligned in time. Standardization converts parameters with different dimensions into a unified dimensionless value; Normalization scales numerical values ​​to the range required by the model input. The preprocessed data is stored in memory for subsequent feature extraction and model inference.

5. The edge computing-based health monitoring system for transformer substations according to claim 3, characterized in that, During feature extraction, appropriate feature extraction methods are used for different types of data, as detailed below: Sliding window statistics and sequence rate of change are used to analyze the time-series characteristics of electrical parameters; One-dimensional convolutional neural network (1DCNN) is used to extract frequency domain features for harmonic characteristics; The trend slope and temperature rise rate are calculated based on the oil temperature / oil level trend characteristics. Extracting statistical features from environmental parameter characteristics; Temperature distribution features are extracted from image features.

6. The edge computing-based health monitoring system for transformer substations according to claim 3, characterized in that, The health assessment model uses a simplified multilayer perceptron model, which has already learned the mapping relationship between features and health indices during the training phase. The multilayer perceptron model outputs the current health index (HI) of the station area and the health scores of each subsystem.

7. The edge computing-based health monitoring system for transformer substations according to claim 6, characterized in that, The Health Index (HI) is calculated using a weighted composite method: HI=100-( ×Voltage imbalance rate+ ×Harmonic distortion rate+ ×Number of times oil temperature exceeded limit+ ×Number of low fuel level alarms+ ×Environmental anomaly rate… ) in: arrive These are the weighting coefficients for each indicator. This value is determined by training a meta-model, which can be optimized based on expert experience and historical data. Edge computing terminal devices calculate health indices in real time and classify health status into different levels based on preset thresholds.

8. The edge computing-based health monitoring system for transformer substations according to claim 7, characterized in that, Edge computing terminal devices perform anomaly detection based on health assessment results; When the health index is lower than a certain threshold or the score of a certain subsystem is lower than the corresponding threshold, it is determined that there is an abnormality or potential fault risk in the area, and the terminal immediately triggers the local alarm mechanism. The local alarm mechanism includes: on the one hand, emitting sound and light signals through the built-in sound and light alarm to alert on-site personnel; on the other hand, sending alarm information to the pre-set mobile phones of maintenance personnel or monitoring platforms via SMS or network communication, including the station area number, alarm type, time of occurrence, and current health index. Alarm events are also recorded in local logs and databases to save key data and characteristics at the time, so as to facilitate subsequent analysis of the cause of the failure and optimization of model parameters; Edge computing terminal devices upload data to the main platform according to a set period when important events occur; the uploaded data includes: fragments of raw sampled data, extracted feature vectors, calculated health index time series, and alarm records; After receiving the data, the main station platform first stores it in the database for long-term preservation, forming a historical archive of the station's operation. Subsequently, the main station platform's data processing module further cleans and integrates the data, and performs feature engineering to generate input suitable for model training.

9. A transformer area health monitoring system based on edge computing according to claim 8, characterized in that, The main platform deploys various machine learning models and analysis tools to achieve in-depth integrated analysis of the health status of the transformer substations: Health trend prediction: Use long short-term memory networks or Transformer models to train historical health index sequences and learn their temporal evolution patterns; A well-trained model can predict health index trends over the next few hours to days, generating trend prediction curves. Fault Risk Assessment: The XGBoost ensemble learning model is used, combined with historical fault case data, to train the multi-source features of the transformer area and output the probability or risk level of a fault occurring in the future. The XGBoost model can handle high-dimensional features and nonlinear relationships, and accurately assess the impact of various factors on fault risk. The main station platform can set different risk levels and automatically determine the risk level of the transformer area based on the model output for reference in operation and maintenance decisions. Periodic anomaly identification: Using time series analysis and pattern recognition techniques, detect whether there are periodic anomalies in the operation data of the transformer area; Operation and maintenance decision support: The main platform integrates the above analysis results and generates operation and maintenance suggestions. These suggestions are provided to operation and maintenance managers through reports or visualization interfaces as a basis for formulating maintenance plans and optimizing operation strategies.

10. A method for detecting the health of transformer substations based on edge computing, characterized in that, Includes the following steps: Edge computing terminal devices are deployed on-site in the distribution area to collect and preprocess multi-source operational data in real time; Features are extracted from the edge side and a health index is calculated using an MLP model; A local alarm is triggered when health indicators are abnormal. Upload key data and health assessment results to the main platform; The main platform uses machine learning models to predict health trends, assess fault risks, and identify periodic anomalies, and generates operation and maintenance suggestions. The main platform distributes the analysis results to edge computing terminal devices to achieve edge-cloud collaboration.

Images