Transformer monitoring method, device and equipment based on digital twinning and medium

Abstract

The invention relates to a transformer monitoring method and device based on digital twinning, equipment and a medium. The method comprises the following steps: acquiring operation state data of a transformer through a physical entity of the transformer; reasoning the operation state data of the transformer through the lightweight twin deployed in the edge side subsystem to obtain a state score of the transformer; determining a monitoring processing mode according to the state score; wherein the monitoring processing mode comprises sending a diagnosis request to the cloud subsystem, confirming a transmission interaction mode with the cloud subsystem through a bandwidth, and executing any one of preset emergency actions; wherein deep twin bodies are deployed in the cloud subsystem, and the deep twin bodies are used for determining the diagnosis result of the transformer. By adopting the method, the efficiency and the accuracy can be improved under the condition of low delay of transformer monitoring through a differential strategy of hierarchical collaboration of deep twinborn bodies of the edges, cloud lightweight twinborn bodies and state scoring.

Description

Technical Field

[0001] This application relates to the field of digital twin technology, and in particular to a transformer monitoring method, apparatus, computer equipment, computer-readable storage medium, and computer program product based on digital twins. Background Technology

[0002] As a core component of the power grid, the health status of transformers directly affects the safe and stable operation of the grid. Digital twin technology, by constructing a virtual mapping of physical entities, provides a new paradigm for real-time monitoring, condition assessment, and predictive maintenance of transformers. However, existing transformer digital twin solutions mainly fall into two technical categories: purely cloud-based digital twin solutions and purely edge-based digital twin solutions.

[0003] The pure cloud-based digital twin solution involves uploading all transformer sensor data to the cloud, where a high-fidelity model is built and run. The advantage of this solution is its powerful computing capabilities, enabling the execution of complex physical models and AI algorithms, resulting in more accurate transformer monitoring results.

[0004] The pure edge digital twin solution completes all computation and decision-making within the smart gateway located locally on the transformer. The advantages of this solution are fast response time and low network dependency.

[0005] However, current pure cloud-based digital twin solutions suffer from high latency, high bandwidth pressure, and high network dependence, while pure edge-based digital twin solutions cannot support complex high-fidelity models, resulting in inaccurate monitoring results of transformers and an inability to update and iterate digital twins. Summary of the Invention

[0006] Therefore, it is necessary to provide a transformer monitoring method, device, computer equipment, computer-readable storage medium, and computer program product based on digital twins that can improve the accuracy of transformer monitoring while reducing the impact of data transmission latency, in order to address the above-mentioned technical problems.

[0007] Firstly, this application provides a transformer monitoring method based on digital twins, comprising:

[0008] By examining the physical entity of the transformer, we can obtain data on its operating status.

[0009] The transformer's status score is obtained by reasoning about the transformer's operating status data using a lightweight twin deployed in the edge-side subsystem.

[0010] The monitoring and handling method is determined based on the status score; the monitoring and handling method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission and interaction method with the cloud subsystem through bandwidth, or executing a preset emergency action; wherein, a deep twin is deployed in the cloud subsystem, and the deep twin is used to determine the diagnostic results of the transformer.

[0011] In one embodiment, determining the monitoring result based on the status score includes:

[0012] If the status score is in the first range, the transformer is determined to be in normal condition, and the network bandwidth is obtained.

[0013] If the network bandwidth is greater than the first preset bandwidth threshold, the pre-processed transformer operating status data will be sent to the cloud subsystem.

[0014] If the network bandwidth is less than the first preset bandwidth threshold and the network bandwidth is greater than the second preset bandwidth threshold, the feature data will be sent to the cloud subsystem; wherein, the feature data is obtained by a lightweight twin extracting features from the preprocessed transformer operating status data;

[0015] If the network bandwidth is less than the second preset bandwidth threshold, the transformer operating status data will be cached locally, and a heartbeat connection with the cloud subsystem will be maintained.

[0016] In one embodiment, the monitoring result is confirmed based on the range of the status score, including:

[0017] If the status score is in the second range, the transformer is determined to be in a warning state. A diagnostic request is then sent to the cloud subsystem so that the deep twin in the cloud subsystem can obtain diagnostic suggestions based on the diagnostic request. The diagnostic request includes feature data corresponding to the transformer's operating status data and the device context.

[0018] Receive diagnostic suggestions from the cloud subsystem and perform transformer maintenance based on the suggestions.

[0019] In one embodiment, the monitoring result is confirmed based on the range of the status score, including:

[0020] If the status score is in the third range, the transformer is determined to be in an emergency state. Then, the preset emergency action is executed through the control interface, and the target data is uploaded to the cloud subsystem through a preset priority channel. The target data includes at least the transformer operating status data. The preset priority channel is the highest priority channel between the edge subsystem and the cloud subsystem.

[0021] In one embodiment, the lightweight twin is obtained by model distillation of a deep twin;

[0022] The method also includes:

[0023] Receive updates to the lightweight twin or parameter update packages sent by the cloud subsystem, and update the local lightweight twin.

[0024] In one embodiment, the feature data is obtained in the following manner:

[0025] The transformer operating status data is preprocessed to obtain a time-series data stream in a unified format; the preprocessing includes data cleaning, alignment and standardization.

[0026] Feature data is obtained by extracting features from time-series data streams using lightweight twins.

[0027] Secondly, this application also provides a transformer monitoring device based on digital twins, comprising:

[0028] The acquisition module is used to obtain transformer operating status data through the physical entity of the transformer;

[0029] The module is used to infer the transformer's operating status data through a lightweight twin deployed in the edge-side subsystem, and obtain the transformer's status score.

[0030] The monitoring and processing module is used to determine the monitoring and processing method based on the status score. The monitoring and processing method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission and interaction method with the cloud subsystem through bandwidth, or executing a preset emergency action. The cloud subsystem is equipped with a deep twin, which is used to determine the diagnostic results of the transformer.

[0031] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the steps of the method described above.

[0032] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0033] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0034] The aforementioned transformer monitoring method, device, computer equipment, computer-readable storage medium, and computer program product based on digital twins acquire transformer operating status data through the physical entity of the transformer; infer the transformer operating status data using a lightweight twin deployed in the edge-side subsystem to obtain a transformer status score; and determine the monitoring processing method based on the status score. The monitoring processing method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission interaction method with the cloud subsystem through bandwidth, or executing a preset emergency action. A deep twin is deployed in the cloud subsystem to determine the transformer's diagnostic results. By focusing on transformer monitoring and rapid repair when a transformer fault occurs through the lightweight twin in the edge-side subsystem, and by using the deep twin deployed in the cloud subsystem to confirm more accurate monitoring results, the system effectively solves the problem of high latency preventing timely identification of transformer status and addresses the inaccuracy of transformer monitoring by digital twins solely at the edge side. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is an application environment diagram of a transformer monitoring method based on digital twins in one embodiment;

[0037] Figure 2 This is a flowchart illustrating a transformer monitoring method based on digital twins in one embodiment;

[0038] Figure 3 A flowchart illustrating another transformer monitoring method based on digital twins is provided.

[0039] Figure 4 This is a flowchart illustrating the steps for confirming monitoring results in one embodiment;

[0040] Figure 5 A flowchart illustrating the steps for confirming monitoring results when the transformer is in a normal state is provided.

[0041] Figure 6 A flowchart illustrating the steps for confirming monitoring results when a transformer is in an early warning state is provided.

[0042] Figure 7This is a structural block diagram of a transformer monitoring device based on digital twins in one embodiment;

[0043] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] The transformer monitoring method based on digital twins provided in this application can be applied to, for example... Figure 1 In the application environment shown, the edge subsystem 101 communicates with the cloud subsystem 102 via a communication network. The edge subsystem 101 acquires transformer operating status data through the physical entity of the transformer; the edge subsystem 101 uses a lightweight twin deployed in the edge subsystem to infer the transformer operating status data and obtain a transformer status score; the edge subsystem 101 determines the monitoring and processing method based on the status score; the monitoring and processing method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission interaction method with the cloud subsystem 102 through bandwidth, or executing a preset emergency action; the cloud subsystem 102 deploys a deep twin, which is used to determine the diagnostic results of the transformer.

[0046] In one exemplary embodiment, such as Figure 2 The diagram illustrates a flowchart of a transformer monitoring method based on digital twins, demonstrating its application in... Figure 1 The following description is based on the edge subsystem 101, including steps 201 to 203. Wherein:

[0047] Step 201: Obtain transformer operating status data through the physical entity of the transformer;

[0048] Among them, the health status of transformers, as core equipment of the power grid, is directly related to the safe and stable operation of the power grid.

[0049] The physical entity of a transformer is equipped with various sensors, which may include oil chromatography sensors, vibration sensors, temperature sensors, current / voltage transformers, ultra-high frequency partial discharge sensors, etc.

[0050] Transformer operating status data can also be referred to as multi-source heterogeneous data, which refers to data from different sensors or systems. Transformer operating status data includes electrical quantities such as voltage, current, active power, reactive power, and frequency; dissolved gas analysis in transformer oil; temperature; partial discharge; vibration data; noise data; oil level; and core grounding data. The characteristic gas content of dissolved gases in transformer oil (also known as insulating oil) can include: H2 (hydrogen), CH4 (methane), C2H2 (ethane), C2H4 (ethylene), C2H6 (acetylene), CO (carbon monoxide), and CO2 (carbon dioxide). Temperature can include top oil temperature, bottom oil temperature, and winding temperature. Partial discharge includes high-frequency, ultra-high-frequency, and ultrasonic discharge. Vibration data can include core vibration and tank vibration.

[0051] Vibration data can be called vibration spectrum, which is a form of vibration data. An abnormal vibration spectrum indicates abnormal vibration of the transformer (tank or core, etc.) and is a judgment on the operating status of the transformer.

[0052] Furthermore, the insulating oil and solid insulating materials (paper, cardboard) inside a transformer are stable under normal operating conditions. However, when early faults occur inside the transformer (such as thermal or electrical effects), the chemical molecular bonds of the insulating oil and insulating paper are broken, producing various characteristic gases that dissolve in the transformer oil. Therefore, different types and degrees of transformer faults result in differences in the types, concentrations, proportions, and gas production rates (gradients of change) of the gases produced. Thus, analyzing the dissolved gases in the transformer oil can help determine the type, location, severity, and development trend of transformer faults.

[0053] For example, transformer operating status data can be obtained through various sensors on the physical entity of the transformer.

[0054] Step 202: The transformer's operating status data is inferred by a lightweight twin deployed in the edge-side subsystem to obtain the transformer's status score.

[0055] The edge-side subsystem is deployed within the edge smart gateway of the transformer body. The edge-side subsystem may include a lightweight twin, a data acquisition and preprocessing module, and an edge communication proxy. The data acquisition and preprocessing module is responsible for high-speed acquisition of raw data from sensors, and for data conditioning, cleaning, formatting, and time alignment. The edge communication proxy is responsible for communicating with the cloud subsystem and integrates a dynamic bandwidth sensing module. The dynamic bandwidth sensing module is used to obtain real-time bandwidth.

[0056] A lightweight twin is a pruned and quantized lightweight AI (Artificial Intelligence) model. For example, a lightweight twin can be a simplified version of an LSTM (Long Short-Term Memory) predictive model, a CNN (Convolutional Neural Network) diagnostic model, or an empirical model library. Its main functions are: High-frequency data filtering: Running in millisecond cycles, it performs real-time analysis of pre-processed data to identify abnormal signs. Millisecond-cycle operation: When a serious anomaly is detected, it can directly execute preset emergency operations through the control interface without waiting for cloud commands, such as starting a backup cooler or issuing a trip signal. Abnormal signs can include sudden changes in characteristic gas content, abnormal vibration spectra, and an increase in partial discharge pulses. Serious anomalies can include a sharp increase in acetylene and excessive partial discharge.

[0057] The status score is a comprehensive score derived by a lightweight twin from preprocessed transformer operating status data, characterizing the health level or abnormal risk of the transformer's current operating status. The scoring criteria are based on model inference, combining multiple features (such as gas content, vibration data, partial discharge, etc.) to calculate a numerical value.

[0058] For example, a transformer status score can be obtained by reasoning from various transformer operating status data using a lightweight twin.

[0059] Step 203: Determine the monitoring and processing method based on the status score; wherein, the monitoring and processing method includes any one of sending a diagnostic request to the cloud subsystem, confirming the transmission and interaction method with the cloud subsystem through bandwidth, or executing a preset emergency action; wherein, a deep twin is deployed in the cloud subsystem, and the deep twin is used to determine the diagnostic results of the transformer.

[0060] Among them, a deep twin is a high-precision digital twin model, which can be a multi-physics coupling model, or an electromagnetic-thermal-fluid coupling simulation model built based on the transformer's geometry and material properties; it can also be a deep AI diagnostic model, based on a complex deep neural network (such as Transformer, graph neural network) trained on massive historical data, used for the accurate identification, location and life prediction of transformer fault types.

[0061] The cloud subsystem is deployed on a cloud computing platform and includes: a deep twin, a model training and iteration engine, and a cloud-based collaborative management platform. The model training and iteration engine utilizes historical and real-time data from multiple transformers collected by the cloud subsystem to continuously train and optimize the parameters of the deep twin. The cloud-based collaborative management platform is responsible for receiving data and events reported from the edge and scheduling cloud computing resources.

[0062] For example, the edge monitoring subsystem determining the monitoring processing method based on the status score may include: when the status score is in the first range, indicating the transformer is in a normal state, the network bandwidth is acquired, and the edge monitoring subsystem confirms the data to be sent to the cloud subsystem based on the network bandwidth. When the status score is in the second range, indicating the transformer is in an early warning state, a diagnostic request is sent to the cloud subsystem, so that the deep twin in the cloud subsystem can obtain diagnostic suggestions and parameter update packages based on the diagnostic request. When the status score is in the third range, a preset emergency action is executed through the control interface, and the emergency event, action record, and related instantaneous data are reported to the cloud subsystem through the highest priority channel for post-event analysis and auditing.

[0063] In the aforementioned transformer monitoring method based on digital twins, transformer operating status data is obtained through the physical entity of the transformer. A lightweight twin deployed in the edge-side subsystem infers the transformer's operating status data to obtain a transformer status score. The monitoring processing method is determined based on the status score. This monitoring processing method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission interaction method with the cloud subsystem via bandwidth, or executing a preset emergency action. A deep twin is deployed in the cloud subsystem to determine the transformer's diagnostic results. By focusing on transformer monitoring and rapid repair when a fault occurs through the lightweight twin in the edge-side subsystem, and by using the deep twin deployed in the cloud subsystem to confirm more accurate monitoring results, this method not only solves the problem of high latency preventing timely identification of transformer status but also addresses the inaccuracy of transformer monitoring by digital twins solely on the edge side.

[0064] Figure 3 Another flowchart of a transformer monitoring method based on digital twins is provided, such as... Figure 3 As shown, it includes:

[0065] S100: System initialization and deployment.

[0066] An initial high-fidelity deep twin (equivalent to the deep twin in this application) is deployed in the cloud (equivalent to the cloud subsystem of this application), and the model is pre-trained based on massive historical data.

[0067] A general, lightweight, real-time twin is derived from a high-fidelity model using model distillation techniques.

[0068] Deploy this general lightweight model into the edge smart gateway of the target transformer.

[0069] Configure the warning threshold Twarning and the emergency action threshold Tcritical on the edge side.

[0070] S200: Continuous data perception and preprocessing on the edge side (equivalent to the edge-side subsystem in this application).

[0071] The edge side synchronously collects the raw data of all sensors at a high frequency (e.g., 1kHz).

[0072] Perform data cleaning, alignment, and standardization to form a time-series data stream in a unified format.

[0073] Perform lightweight feature extraction in real time, such as calculating vibration spectra, gas content change gradients, etc.

[0074] S300: Real-time inference and local decision-making on the edge side.

[0075] The lightweight real-time twin performs inference on the preprocessed feature data at a fixed period (e.g., 100ms) and calculates the current state score S.

[0076] Judgment logic:

[0077] If S≥Twarning, the state is normal, and transfer to S400: Dynamic bandwidth-aware data transmission (under normal circumstances).

[0078] The edge communication agent obtains the current network bandwidth B in real time.

[0079] If B>B1, adopt the full-volume mode and pack and upload the data processed by S202 to the cloud. If B2<B≤B1, adopt the feature mode and only upload the extracted feature vectors. If B≤B2, adopt the extreme mode and only maintain the heartbeat connection and cache all data.

[0080] If Tcritical≤S<Twarning, transfer to S500: Adaptive model switching mechanism (under warning state).

[0081] The edge side immediately sends a "deep diagnosis request" to the cloud collaborative management platform, and the request is attached with high-frequency feature data segments and device context information (such as the current load rate) in the recent period (e.g., 5 minutes).

[0082] The cloud platform receives the request and starts a high-priority computing task.

[0083] The high-fidelity deep twin on the cloud performs in-depth analysis on the received data segments, and combines with the full-volume historical data to give accurate diagnostic conclusions (such as "medium-energy discharge, near the tap changer, confidence level 90%") and predictive suggestions.

[0084] Model distillation and personalized update:

[0085] The cloud engine analyzes the decision path of the high-fidelity model in this diagnosis. "Refine" this path into a set of adjustments to the edge lightweight model, such as: increasing the weight of the feature of the C2H2 / H2 ratio; adjusting a support vector of the SVM classifier; or updating the bias term of a certain layer of the micro neural network. Generate a lightweight model parameter incremental update package.

[0086] The cloud side sends the response message including the diagnostic report and the model update package to the edge side.

[0087] After receiving the message, the edge side first performs local alarms and logging of the diagnostic report. Subsequently, on the premise of ensuring system stability, it hot-updates some parameters of its lightweight real-time twin, so as to improve its recognition ability for similar faults in the future.

[0088] If S < Tcritical, transfer to S600: millimeter-level edge emergency response (in an emergency).

[0089] Without confirmation from the cloud side, the edge side immediately executes actions according to the preset security logic.

[0090] At the same time, the edge side reports the emergency event, action records, and relevant instantaneous data to the cloud through the highest-priority channel for post-event analysis and auditing.

[0091] S700: Cloud data reception and continuous iteration and evolution of the cloud model.

[0092] The cloud continuously receives data (full volume or features) uploaded from all edge nodes.

[0093] The model training engine periodically (such as daily) retrains the high-fidelity deep twin using the newly accumulated data to achieve version iteration of the model (such as upgrading from V1.0 to V1.1).

[0094] When the cloud model completes a major version upgrade, it can re-trigger the steps of S100 to send a new basic lightweight model to the edge side to complete the large-loop evolution of the entire system.

[0095] In an exemplary embodiment, as Figure 4 shown, Figure 4 Provide a step for confirming the monitoring result. The step of confirming the monitoring result includes steps 401 to 404. Among them:

[0096] Step 401, if the status score is within the first range, determine that the transformer is in a normal state, and then obtain the network bandwidth.

[0097] The state score S is a comprehensive score derived by the lightweight twin from the preprocessed transformer operating status data, representing the health level or abnormal risk of the transformer's current operating status. The scoring criteria can be calculated based on model inference, combined with multiple features (such as gas content, vibration data, partial discharge, etc.). The state score includes a first range, a second range, and a third range. The first range represents state scores greater than Twarning, the second range represents state scores between Twarning and Tcritical, and the third range represents state scores less than Twarning. Twarning and Tcritical are thresholds configured during system initialization, obtained based on historical data and model training.

[0098] S can be a score from 0 to 100 or a similar indicator, with lower values ​​indicating worse transformer condition (higher risk of anomalies). For example, S is a risk score output by a lightweight twin model based on transformer operating status data.

[0099] Network bandwidth is a core indicator for measuring signal transmission capability or data transmission rate.

[0100] For example, by inputting the transformer's operating status into the lightweight twin, a status score is obtained. When the status score is greater than Twarning, the transformer is confirmed to be in a normal state, and the real-time bandwidth is obtained.

[0101] Step 402: If the network bandwidth is greater than the first preset bandwidth threshold, the preprocessed transformer operating status data is sent to the cloud subsystem.

[0102] Among them, the first preset bandwidth can be denoted as B1, the second preset bandwidth can be denoted as B2, and the network bandwidth can be denoted as B. B1 and B2 are bandwidth thresholds used in the dynamic bandwidth perception module of the edge-side subsystem. They are preset and obtained by the edge-side subsystem during system initialization. After the subsequent dynamic bandwidth perception module runs, they are obtained based on network performance testing. In one embodiment, assume a transformer monitoring scenario: a transformer equipped with sensors such as oil chromatograph, vibration, and temperature, and carrying an edge intelligent device. The dynamic bandwidth perception module confirms the bandwidth requirements of full-volume data (data generated by various monitoring sensors, as well as data such as transformer specifications, models, and numbers), feature data (extracted feature vectors), instructions, etc. Then, based on the requirements, considering the overhead of the TCP (Transmission Control Protocol) / IP (Internet Protocol) protocol and the margin required for network fluctuations, B1 (theoretical value) is calculated. Confirm the bandwidth required to maintain the most core "model evolution and emergency communication", and similarly consider the protocol overhead and margin to calculate B2 (theoretical value). Then the dynamic bandwidth perception module obtains the current network bandwidth in real time for network performance testing, analyzes the network bandwidth distribution in this area through network performance, and combines the actual situation to fine-tune the calculated B1 and B2 values (theoretical values) by the dynamic bandwidth perception module. So that the system can provide a good full-volume data transmission experience for the vast majority of the time.

[0103] Exemplarily, if B > B1, it indicates that the full-volume mode is enabled. The full-volume mode indicates that the current network bandwidth state is good and a large amount of data can be sent. Therefore, the preprocessed transformer operating state data is sent to the cloud terminal subsystem.

[0104] Step 403, if the network bandwidth is less than the first preset bandwidth threshold and greater than the second preset bandwidth threshold, the feature data is sent to the cloud terminal subsystem; among them, the feature data is obtained by the lightweight twin performing feature extraction on the preprocessed transformer operating state data.

[0105] Among them, the lightweight twin performs a feature extraction operation on the preprocessed transformer operating state data to obtain the feature data.

[0106] Exemplarily, if B2 < B ≤ B1, it indicates that the current network can support small-data volume transmission, and small-data volume transmission will not cause high latency, which means the feature mode is adopted. The feature mode can indicate that the feature data is sent to the cloud terminal subsystem. Exemplarily, the data volume of the feature data is less than the preprocessed transformer operating state data.

[0107] Step 404, if the network bandwidth is less than the second preset bandwidth threshold, the transformer operating state data is cached locally and the heartbeat connection with the cloud terminal subsystem is maintained.

[0108] Among them, heartbeat connection is a mechanism in network communication that uses the periodic sending of lightweight data packets (heartbeat packets) to detect whether the connection is alive and maintain the connection state. It is widely used in long-connection, distributed systems and instant messaging fields.

[0109] For example, if B≤B2, then the extreme mode is adopted, only the heartbeat connection is maintained, and all data is cached. All data may include transformer operating status data and characteristic data.

[0110] Figure 5 A flowchart illustrating the steps for confirming monitoring results when the transformer is in a normal state is provided, such as... Figure 5 As shown, the available network bandwidth B is detected in real time. If B2 < B ≤ B1, it is in feature mode, and only feature data extracted by the edge subsystem is uploaded. If B ≤ B2, it is in extreme mode, and the edge subsystem only maintains a heartbeat connection with the cloud subsystem. If B is greater than B1, it is in full mode, and all transformer operating status data or high-density data is uploaded.

[0111] In one embodiment, confirming the monitoring result based on the range of the status score includes:

[0112] If the status score is in the second range, the transformer is determined to be in a warning state. A diagnostic request is then sent to the cloud subsystem so that the deep twin in the cloud subsystem can obtain diagnostic suggestions based on the diagnostic request. The diagnostic request includes feature data corresponding to the transformer's operating status data and the device context.

[0113] Receive diagnostic suggestions from the cloud subsystem and perform transformer maintenance based on the suggestions.

[0114] The diagnostic request may include high-frequency characteristic data and device context information from a recent period (e.g., 5 minutes). The device context information may include the current load rate, and the high-frequency characteristic data may include transformer status and operating data.

[0115] For example, diagnostic recommendations include a diagnostic report and a parameter update package. The diagnostic recommendations may include: a medium-energy discharge located near a tap changer, with a 90% confidence level. The parameter update package is used to update the lightweight twin.

[0116] The parameter update package may include "refining" the decision path into a set of adjustments to the edge lightweight model (which can also be called the lightweight twin), such as increasing the weight of the feature "C2H2 / H2 ratio"; adjusting a support vector of the SVM (Support Vector Machine) classifier; or updating the bias term of a certain layer of the micro neural network in the lightweight twin. Thus, a lightweight model parameter incremental update package (which can also be called the parameter update package) is generated. Among them, the decision path refers to the complete decision-making process from the input data to the model output result, which shows how the model gradually derives the final judgment based on the input data. The input data can refer to the transformer operation status data.

[0117] Figure 6 A flow schematic diagram of the steps for confirming the monitoring result when the transformer is in the early warning state is provided, including:

[0118] If Tcritical ≤ S < Twarning, it is confirmed that the transformer is in the early warning state, and the adaptive model switching mode is triggered. The adaptive model switching mode is: the edge-side subsystem immediately sends a diagnostic request to the cloud collaborative management platform in the cloud terminal subsystem. The diagnostic request is accompanied by high-frequency feature data and device context information for a recent period (such as 5 minutes).

[0119] The cloud-side subsystem receives the diagnostic request and starts a high-priority computing task according to the diagnostic request. The deep twin of the cloud-side subsystem performs in-depth analysis on the received feature data, and combines the high-frequency feature data and device context information to obtain an accurate diagnostic report and parameter update package. In one embodiment, the diagnostic report may include "medium-energy discharge, near the tap changer, confidence level 90%" and predictive suggestions. The cloud engine analyzes the decision path of the high-fidelity model (which can also be called the deep twin) in this diagnosis. "Refines" the decision path into a set of adjustments to the edge lightweight model (which can also be called the lightweight twin) to obtain the parameter update package. The parameter update package can be: increasing the weight of the feature "C2H2 / H2 ratio"; adjusting a support vector of the SVM (Support Vector Machine) classifier; or updating the bias term of a certain layer of the micro neural network. A lightweight model parameter update package is generated.

[0120] The cloud-side subsystem sends the diagnostic suggestions to the edge-side subsystem. Among them, the diagnostic suggestions include the parameter update package and the diagnostic report. <00002​After the edge-side subsystem receives the diagnostic suggestions, it first performs processing operations according to the diagnostic report. Subsequently, on the premise of ensuring system stability, it hot-updates some parameters of its lightweight real-time twin according to the parameter update package, so as to improve its ability to identify similar faults in the future.

[0122] Through the data transmission strategy of transformer status grading and network bandwidth adaptation, combined with the lightweight twin feature extraction and heartbeat connection keep-alive mechanism, the intelligent, efficient and reliable transmission of transformer operation status data is achieved.

[0123] In one embodiment, according to the range where the status score is located, the monitoring result is confirmed, including:

[0124] If the status score is in the third range, it is determined that the transformer is in an emergency state. Then, a preset emergency action is executed through the control interface and the target data is uploaded to the cloud terminal subsystem through a preset priority channel; where the target data at least includes the transformer operation status data; where the preset priority channel is the highest priority channel between the edge-side subsystem and the cloud terminal subsystem.

[0125] Among them, executing a preset emergency action through the control interface can refer to the execution of actions by the safety logic, which can specifically include starting the standby cooler and sending a trip signal.

[0126] Exemplarily, if S < Tcritical, the edge-side subsystem immediately executes actions according to the preset safety logic without confirmation by the cloud terminal subsystem. At the same time, the edge-side reports this emergency event, action records and related instantaneous data to the cloud through the highest priority channel for post-event analysis and auditing.

[0127] In one embodiment, where the lightweight twin is obtained by performing model distillation on the deep twin; the method further includes:

[0128] Receive the updated lightweight twin or parameter update package sent by the cloud terminal subsystem and update the local lightweight twin.

[0129] Model distillation, also known as knowledge distillation, is a technique for transferring knowledge from a complex and large teacher model to a lightweight student model. The teacher model is a pre-trained, high-performance large model, such as ResNet (Residual Network) or GPT (Generative Pre-trained Transformer) series. The student model is a simpler, more computationally efficient lightweight model, such as MobileNet or TinyBERT (Bidirectional Encoder Representations from Transformers).

[0130] For example, when the model in the deep twin of the cloud-side subsystem completes a major version upgrade, it sends an updated lightweight twin to the cloud-side subsystem. The cloud-side subsystem continuously receives data (fully quantized features) uploaded from all edge nodes. The model training engine periodically (e.g., daily) retrains the deep twin using the newly accumulated data to achieve model version iteration, which can be from V1.0 to V1.1. In one embodiment, a major version upgrade is defined as a change in the model's content by a preset number of changes, which can be 50%, equivalent to a 50% change in content compared to the previous version. The following operations can be performed again: deploy the initial deep twin in the cloud-side subsystem. This model (i.e., the deep twin) is pre-trained based on massive amounts of historical data. Then, a new basic lightweight model (equivalent to a lightweight twin) is distributed to the edge-side subsystem, completing the large-scale evolution of the entire system. When the transformer is in an alert state, it receives parameter update packages sent by the cloud subsystem. In another embodiment, a preset update time can also be set, and when the preset update time is reached, a parameter update package sent by the cloud subsystem can be received.

[0131] In one embodiment, the feature data is obtained in the following manner:

[0132] The transformer operating status data is preprocessed to obtain a time-series data stream in a unified format; the preprocessing includes data cleaning, alignment and standardization.

[0133] Feature data is obtained by extracting features from time-series data streams using lightweight twins.

[0134] Data cleaning is the core step in data preprocessing. It involves identifying and processing dirty data (missing, anomalous, duplicate, inconsistent, redundant, etc.) in a dataset to obtain high-quality, consistent, and usable data. Its core objective is to improve data quality.

[0135] Alignment refers to the process of unifying multi-source, multi-dimensional, and multi-format objects under the same benchmark / standard, eliminating deviations and inconsistencies, and ensuring the accuracy and effectiveness of subsequent operations. Multi-format objects include data, features, model outputs, system interfaces, etc.

[0136] Standardization refers to mapping feature data with different dimensions and value ranges to a unified numerical scale through mathematical transformation, eliminating the differences in dimensions and numerical ranges between features, and enabling the model to learn the contribution of each feature fairly.

[0137] For example, by performing data cleaning, alignment and standardization on transformer operating status data, a time-series data stream with a unified format is obtained. Then, feature data is obtained by extracting features from the time-series data stream using a lightweight twin.

[0138] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0139] Based on the same inventive concept, this application also provides a digital twin-based transformer monitoring device for implementing the aforementioned digital twin-based transformer monitoring method. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations of one or more embodiments of the digital twin-based transformer monitoring device provided below can be found in the limitations of the digital twin-based transformer monitoring method described above, and will not be repeated here.

[0140] In one exemplary embodiment, such as Figure 7As shown, a transformer monitoring device based on digital twins is provided, including: an acquisition module 701, a processing module 702, and a monitoring and processing module 703, wherein:

[0141] The acquisition module 701 is used to acquire transformer operating status data through the physical entity of the transformer;

[0142] Module 702 is obtained, which is used to infer the transformer operating status data through a lightweight twin deployed in the edge-side subsystem to obtain the transformer status score;

[0143] The monitoring and processing module 703 is used to determine the monitoring and processing method based on the status score. The monitoring and processing method includes any one of the following: sending a diagnostic request to the cloud subsystem, confirming the transmission and interaction method with the cloud subsystem through bandwidth, or executing a preset emergency action. A deep twin is deployed in the cloud subsystem, and the deep twin is used to determine the diagnostic results of the transformer.

[0144] In one exemplary embodiment, the monitoring processing module 703 can also be used to:

[0145] If the status score is in the first range, the transformer is determined to be in normal condition, and the network bandwidth is obtained.

[0146] If the network bandwidth is greater than the first preset bandwidth threshold, the pre-processed transformer operating status data will be sent to the cloud subsystem.

[0147] If the network bandwidth is less than the first preset bandwidth threshold and the network bandwidth is greater than the second preset bandwidth threshold, the feature data will be sent to the cloud subsystem; wherein, the feature data is obtained by a lightweight twin extracting features from the preprocessed transformer operating status data;

[0148] If the network bandwidth is less than the second preset bandwidth threshold, the transformer operating status data will be cached locally, and a heartbeat connection with the cloud subsystem will be maintained.

[0149] In one exemplary embodiment, the monitoring processing module 703 can also be used to:

[0150] If the status score is in the second range, the transformer is determined to be in a warning state. A diagnostic request is then sent to the cloud subsystem so that the deep twin in the cloud subsystem can obtain diagnostic suggestions based on the diagnostic request. The diagnostic request includes feature data corresponding to the transformer's operating status data and the device context.

[0151] Receive diagnostic suggestions from the cloud subsystem and perform transformer maintenance based on the suggestions.

[0152] In one exemplary embodiment, the monitoring processing module 703 can also be used to:

[0153] If the status score is in the third range, the transformer is determined to be in an emergency state. Then, the preset emergency action is executed through the control interface, and the target data is uploaded to the cloud subsystem through a preset priority channel. The target data includes at least the transformer operating status data. The preset priority channel is the highest priority channel between the edge subsystem and the cloud subsystem.

[0154] In one exemplary embodiment, the monitoring processing module 703 can also be used to:

[0155] Receive updates to the lightweight twin or parameter update packages sent by the cloud subsystem, and update the local lightweight twin.

[0156] In one exemplary embodiment, the monitoring processing module 703 can also be used to:

[0157] The transformer operating status data is preprocessed to obtain a time-series data stream in a unified format; the preprocessing includes data cleaning, alignment and standardization.

[0158] Feature data is obtained by extracting features from time-series data streams using lightweight twins.

[0159] The modules in the aforementioned digital twin-based transformer monitoring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0160] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores transformer operating status data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a transformer monitoring method based on digital twins.

[0161] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0162] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0163] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0164] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0165] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0166] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0167] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0168] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A transformer monitoring method based on digital twins, characterized in that, Applied to an edge-side subsystem, the method includes: By examining the physical entity of the transformer, we can obtain data on its operating status. The transformer's status score is obtained by reasoning through the transformer's operating status data using a lightweight twin deployed in the edge-side subsystem. The monitoring and processing method is determined based on the status score; wherein the monitoring and processing method includes any one of sending a diagnostic request to the cloud subsystem, confirming the transmission interaction method with the cloud subsystem through bandwidth, and executing a preset emergency action; wherein a deep twin is deployed in the cloud subsystem, and the deep twin is used to determine the diagnostic result of the transformer.

2. The method according to claim 1, characterized in that, Determining the monitoring result based on the status score includes: If the status score is within the first range, the transformer is determined to be in a normal state, and the network bandwidth is obtained. If the network bandwidth is greater than the first preset bandwidth threshold, the preprocessed transformer operating status data will be sent to the cloud subsystem. If the network bandwidth is less than a first preset bandwidth threshold and the network bandwidth is greater than a second preset bandwidth threshold, then the feature data is sent to the cloud subsystem; wherein, the feature data is obtained by the lightweight twin extracting features from the preprocessed transformer operating status data; If the network bandwidth is less than the second preset bandwidth threshold, the transformer operating status data is cached locally, and a heartbeat connection with the cloud subsystem is maintained.

3. The method according to claim 1, characterized in that, The step of confirming the monitoring result based on the range of the status score includes: If the status score is within the second range, the transformer is determined to be in a warning state. Then, a diagnostic request is sent to the cloud subsystem so that the deep twin in the cloud subsystem can obtain diagnostic suggestions based on the diagnostic request. The diagnostic request includes feature data corresponding to the transformer operating status data and device context. The system receives diagnostic suggestions from the cloud subsystem and performs maintenance on the transformer based on these suggestions.

4. The method according to claim 1, characterized in that, The step of confirming the monitoring result based on the range of the status score includes: If the status score is in the third range, the transformer is determined to be in an emergency state. Then, a preset emergency action is executed through the control interface, and the target data is uploaded to the cloud subsystem via a preset priority channel. The target data includes at least the transformer operating status data. The preset priority channel is the highest priority channel between the edge subsystem and the cloud subsystem.

5. The method according to any one of claims 1-4, characterized in that, in, The lightweight twin is obtained by model distillation of the depth twin; The method further includes: Receive the updated lightweight twin or parameter update package sent by the cloud subsystem, and update the local lightweight twin.

6. The method according to claim 2 or 3, characterized in that, The feature data is obtained through the following methods: The transformer operating status data is preprocessed to obtain a time-series data stream in a unified format; wherein, the preprocessing includes data cleaning, alignment and standardization. Feature data is obtained by extracting features from the time-series data stream using the lightweight twin.

7. A transformer monitoring device based on digital twin, characterized in that, The device includes: The acquisition module is used to acquire transformer operating status data through the physical entity of the transformer; The module is used to infer the transformer's operating status data through a lightweight twin deployed in the edge-side subsystem to obtain the transformer's status score; The monitoring and processing module is used to determine the monitoring and processing method based on the status score; wherein, the monitoring and processing method includes any one of sending a diagnostic request to the cloud subsystem, confirming the transmission interaction method with the cloud subsystem through bandwidth, and executing a preset emergency action; wherein, a deep twin is deployed in the cloud subsystem, and the deep twin is used to determine the diagnostic result of the transformer.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images