Ring main unit digital twin self-evolution synchronization method, terminal device, medium and product

Abstract

The application relates to the field of power system automation, and discloses a ring main unit digital twin self-evolution synchronization method, a terminal device, a medium and a product. The method comprises the following steps: obtaining a standardized multi-modal feature set according to multi-modal original data in a ring main unit operation process; obtaining state deviations according to the standardized multi-modal feature set and a local digital twin submodel; synthesizing a global deviation based on the state deviations; determining deviation source features and corresponding digital twin submodel nodes according to the state deviations; triggering hierarchical self-evolution correction of the local digital twin submodel for the digital twin submodel nodes corresponding to the deviation source features according to a comparison relationship between the global deviation and a preset threshold; and driving an edge cloud communication unit to perform hierarchical adaptive synchronization transmission to a cloud platform according to a working condition result of the hierarchical self-evolution correction until a stop instruction is received or a protection shutdown condition is triggered, so that the technical problems of low virtual-real synchronization accuracy of ring main unit multi-modal data, great edge cloud interaction pressure and poor working condition adaptability are solved.

Description

Technical Field

[0001] This application relates to the field of power system automation, and in particular to a method for self-evolution and synchronization of digital twins of ring main units, terminal equipment, media, and products. Background Technology

[0002] Ring main units (RNBs) are critical equipment in power distribution networks, enabling load distribution, fault isolation, and power restoration. Their operational status directly impacts the reliability of the distribution network. Traditional methods for monitoring RNBs often rely on single-source data acquisition, periodic inspections, and reactive maintenance, resulting in incomplete status awareness, poor real-time performance, and weak fault early warning capabilities. In recent years, digital twin technology has been increasingly applied to power equipment operation and maintenance. Typically, a corresponding digital twin model is established on a cloud platform, transmitting various sensor data collected at the edge to the cloud for centralized computation and analysis.

[0003] However, the inventors have found at least the following technical problems in the related technologies: low virtual-real synchronization accuracy of multimodal data in ring main units, high pressure of edge-cloud interaction, and poor adaptability to operating conditions. Summary of the Invention

[0004] One objective of this application is to provide a method, terminal equipment, medium, and product for the self-evolution and synchronization of digital twins of ring main units, at least to solve the technical problems of low virtual-real synchronization accuracy, high edge-cloud interaction pressure, and poor adaptability of operating conditions in related technologies.

[0005] To achieve the above objectives, some embodiments of this application provide the following aspects:

[0006] In a first aspect, some embodiments of this application also provide a method for the self-evolution and synchronization of a ring main unit's digital twin. This method is applied to edge computing nodes and includes: obtaining a standardized multimodal feature set based on multimodal raw data during the ring main unit's operation; the standardized multimodal feature set includes feature values ​​aligned with timestamps and spatial coordinates; performing a virtual-real consistency comparison between the standardized multimodal feature set and the output data of a local digital twin sub-model to obtain state deviations, and synthesizing a global deviation based on each of the state deviations; determining the deviation source features and corresponding twin sub-model nodes based on the state deviations; triggering hierarchical self-evolution correction of the local digital twin sub-model for the twin sub-model nodes corresponding to the deviation source features based on the comparison relationship between the global deviation and a preset threshold; driving the edge-cloud communication unit to perform hierarchical adaptive synchronization transmission to the cloud platform based on the operating conditions of the hierarchical self-evolution correction; and repeatedly executing the above steps until a stop command is received or a protection shutdown condition is triggered.

[0007] Secondly, some embodiments of this application also provide a terminal device, the terminal device comprising: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method described above.

[0008] Thirdly, some embodiments of this application also provide a computer-readable medium having computer program instructions stored thereon, which can be executed by a processor to implement the method described above.

[0009] Fourthly, some embodiments of this application also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method described above.

[0010] Compared with related technologies, the solution provided in this application obtains a standardized feature set by spatiotemporally aligning the original multimodal data. This allows subsequent virtual-to-real comparisons to be based on relatively accurate data, reducing comparison errors caused by data temporal or spatial asynchrony to a certain extent. By performing a consistency comparison between the standardized feature set and the output of the twin model and synthesizing the global deviation, the differences in multiple dimensions can be summarized into a comprehensive value, providing a quantitative basis for judging whether the model needs correction. After locating the deviation source features and the corresponding twin model nodes based on the state deviation, the model parameters can be adjusted in a targeted manner without updating the entire model every time, which helps reduce the computational burden on edge computing nodes. Different levels of self-evolutionary correction are triggered based on the comparison results of the global deviation and a preset threshold, enabling the system to select an appropriate correction method according to the magnitude of the deviation, finding a relatively reasonable trade-off between computational resources and model accuracy. At the same time, in the event of a failure, model updates can be temporarily frozen to prevent the model from being affected by erroneous data, improving system security. Performing hierarchical adaptive synchronous transmission based on the corrected operating conditions allows the system to automatically adjust the frequency and content of data uploads according to the current operating conditions. This reduces unnecessary bandwidth consumption during normal operation while ensuring timely uploading of critical data to the cloud in case of anomalies. Repeating these steps until a stop command is received or protection shutdown conditions are triggered creates a continuous monitoring and dynamic correction loop. This helps the local digital twin model maintain relative consistency between the virtual and real systems as the actual operating status of the ring main unit changes, thus providing a reference for ring main unit status monitoring and maintenance judgment. Attached Figure Description

[0011] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0012] Figure 1 An exemplary flowchart of a ring main unit digital twin self-evolution synchronization method provided for some embodiments of this application;

[0013] Figure 2 A schematic diagram of the overall system architecture in a ring main unit digital twin self-evolution synchronization method provided in some embodiments of this application;

[0014] Figure 3 A schematic diagram of multimodal data fusion at the edge side in a ring main unit digital twin self-evolution synchronization method provided in some embodiments of this application;

[0015] Figure 4 A schematic diagram of edge-cloud collaborative twin synchronization in a ring main unit digital twin self-evolution synchronization method provided in some embodiments of this application;

[0016] Figure 5 This is an exemplary structural diagram of a terminal device provided for some embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] This application relates to a method for the self-evolution and synchronization of a digital twin of a ring main unit, which is applied to edge computing nodes. Please refer to [link to relevant documentation]. Figure 1 The method may include the following steps:

[0019] Step S101: Based on the multimodal raw data during the operation of the ring main unit, a standardized multimodal feature set is obtained; the standardized multimodal feature set includes feature values ​​aligned with timestamps and spatial coordinates.

[0020] Step S102: Perform a virtual-real consistency comparison between the standardized multimodal feature set and the output data of the local digital twin model to obtain the state deviation, and synthesize the global deviation based on each of the state deviations.

[0021] Step S103: Based on the state deviation, determine the deviation source features and the corresponding twin model nodes;

[0022] Step S104: Based on the comparison relationship between the global deviation and the preset threshold, trigger the hierarchical self-evolution correction of the local digital twin model for the twin model node corresponding to the deviation source feature;

[0023] Step S105: Based on the hierarchical self-evolution correction working condition results, drive the edge-cloud communication unit to perform hierarchical adaptive synchronous transmission to the cloud platform.

[0024] Step S106: Repeat the above steps until a stop command is received or the protection shutdown condition is triggered.

[0025] For example, the overall system architecture for implementing the methods provided in the embodiments of this application is described below. Figure 2 This diagram illustrates a system architecture according to an embodiment of this application. The overall system architecture for implementing the method provided in this embodiment may include the following six units, from bottom to top: multimodal perception layer, edge computing layer, digital twin model layer, and edge-cloud collaboration and application layer. The edge-cloud collaboration and application layer further includes an edge-cloud communication unit and cloud platform-related functions.

[0026] See Figure 2 The first layer is the multimodal sensing layer. This layer can include electrical quantity acquisition modules, environmental quantity acquisition modules, mechanical status acquisition modules, and insulation status acquisition modules. The electrical quantity acquisition modules can include sensors such as voltage and current transformers to collect electrical quantity data from the ring main unit; the environmental quantity acquisition modules can include temperature and humidity sensors and SF6 gas sensors to collect environmental quantity data; the mechanical status acquisition modules can include mechanical position sensors to collect mechanical status data; and the insulation status acquisition modules can include partial discharge sensors and vibration sensors to collect insulation status data. All of these acquisition modules can be connected to the edge gateway via RS485 bus, Ethernet, or LoRa communication methods to collect multimodal raw data during the operation of the ring main unit in real time.

[0027] The second layer is the edge computing layer. This layer may include a data preprocessing module, a multimodal feature fusion module, a virtual-real consistency comparison module, and a self-evolution synchronization module. The data preprocessing module performs signal denoising, outlier removal, missing value interpolation, and dimensional normalization on the raw multimodal data. The multimodal feature fusion module performs temporal and spatial alignment on the processed data, outputting a standardized multimodal feature set. The virtual-real consistency comparison module compares the standardized multimodal feature set with the output data of the local digital twin model to obtain state deviations and synthesize a global deviation. The self-evolution synchronization module triggers hierarchical self-evolution correction for the twin model nodes corresponding to the deviation source features based on the comparison relationship between the global deviation and a preset threshold, and drives the edge-cloud communication unit to perform hierarchical adaptive synchronization transmission. The edge computing layer can adopt a heterogeneous architecture combining an ARM processor and a field-programmable gate array (FPGA), with a built-in lightweight artificial intelligence inference module for performing rapid inference and self-evolution correction calculations of the digital twin model at the edge.

[0028] The third layer is the digital twin model layer. This layer can include geometric, electrical, thermal, mechanical, and insulation models, forming a multiphysics coupled model. The digital twin model layer is used to construct the local digital twin sub-model of the ring main unit, supporting the import of 3D computer-aided design models and parametric modeling and multiphysics coupled simulation functions, realizing the virtual mapping of multiphysics characteristics such as the geometric structure, electrical behavior, thermodynamic behavior, mechanical behavior, and insulation behavior of the ring main unit.

[0029] The fourth layer is the edge-cloud collaboration and application layer. This layer can include the edge-cloud communication unit and the cloud platform. The edge-cloud communication unit is responsible for data interaction between edge computing nodes and the cloud platform, supporting multiple transmission modes such as low-frequency periodic synchronization, high-frequency real-time push, and communication interception. It is used to upload data of corresponding levels to the cloud platform according to different operating conditions. The cloud platform can provide functions such as status monitoring, fault early warning, health assessment, and intelligent operation and maintenance. The edge computing layer and the digital twin model layer interact through synchronization and feedback mechanisms, and the digital twin model layer and the edge-cloud collaboration and application layer also interact through synchronization and feedback mechanisms, thereby realizing a global twin closed loop for edge-cloud collaboration.

[0030] In addition, the system may include a human-machine interaction unit. This unit can provide a visual operation interface to display the operating status of the ring main unit, the virtual-to-real comparison results of the digital twin model, and information on the self-evolution and correction process, and supports interactive operations by maintenance personnel.

[0031] The overall system architecture described above can be integrated with power distribution automation systems and operation and maintenance management platforms to form a closed-loop intelligent operation and maintenance system covering multimodal perception, edge computing, digital twin modeling, self-evolution correction, edge-cloud collaboration, and human-machine interaction.

[0032] The following sections will provide a detailed explanation of each of the above steps.

[0033] Specifically, in step S101, the edge computing node can collect multimodal raw data generated during the operation of the ring main unit in real time. The edge computing node can clean, denoise, and normalize the collected raw data to eliminate dimensional differences and noise interference between sensors. Subsequently, the edge computing node can label each processed frame of data with a unified timestamp and combine data from the same moment in chronological order into a feature vector. Simultaneously, based on the actual installation coordinates of each sensor within the ring main unit, the feature vector is mapped to the corresponding spatial coordinate point. After this dual temporal and spatial alignment, the edge computing node can output a standardized multimodal feature set, where each feature value carries both a timestamp and spatial coordinate information.

[0034] Specifically, for step S102, the virtual-real consistency comparison can be achieved in the following way: the edge computing node can compare the measured data contained in the standardized multimodal feature set with the theoretical prediction values ​​output by the locally stored digital twin model point by point. By calculating the difference between the measured value and the predicted value at the same time coordinate and the same spatial coordinate location, the edge computing node can obtain various state deviations.

[0035] Different types of state deviations have varying degrees of ability to characterize the overall operating status of a ring main unit (RNB). For example, under heavy load conditions, current deviations are more sensitive than temperature deviations. To address these characteristics, edge computing nodes can assign different importance weights to different types of state deviations based on the actual operating conditions of the RNB, and then perform a weighted summation of the various state deviations, fusing the deviation information from multiple dimensions into a single global deviation value. This global deviation can comprehensively reflect the overall degree of consistency between the current physical RNB and the digital twin model.

[0036] For step S103, for example, the edge computing node can perform deviation source analysis based on the calculated correlation between various state deviations in the ring main unit equipment topology. By analyzing the spatial proximity and physical coupling relationships between deviations, the edge computing node can determine the core characteristic parameters that cause these deviations, i.e., deviation source characteristics. Furthermore, the edge computing node can locate the corresponding twin sub-model node in the hierarchical digital twin model according to the preset mapping relationship of the deviation source characteristics in the digital twin model, thereby clarifying the specific target of subsequent correction operations.

[0037] For step S104, for example, the edge computing node can compare the calculated global deviation with multiple pre-set thresholds, and trigger different levels of self-evolutionary correction operations based on the comparison results. For the twin model node corresponding to the deviation source feature, the edge computing node can choose to perform parameter-level fine-tuning correction, local network-level incremental training correction, or latching and inference operations under fault mode, depending on the range of the global deviation. These three different levels of correction operations are collectively referred to as hierarchical self-evolutionary correction, which aims to enable the local digital twin model to adaptively evolve its form and parameters according to actual operating conditions.

[0038] For step S105, for example, the edge computing node can drive the edge-cloud communication unit to execute a matching adaptive synchronization transmission strategy based on the operating condition results determined after the hierarchical self-evolution correction. The edge computing node can upload data of different levels to the cloud platform at different frequencies and with different content. By dynamically adjusting the upload frequency, upload content, and upload priority for different operating conditions such as normal, abnormal, and fault conditions, efficient and adaptive synchronization between the edge and cloud is achieved.

[0039] Regarding step S106, as an exemplary startup method, after the system power-on initialization is completed, the edge node calibration is completed, the basic model is successfully loaded, and there is no fault interlocking signal, the edge computing node automatically enters the real-time loop execution state. The edge computing node periodically performs multimodal feature extraction, virtual-real consistency comparison, deviation tracing, hierarchical evolution correction, and hierarchical synchronous transmission processes, forming a continuous self-evolution closed loop until it receives an external stop command or triggers the preset protection shutdown conditions.

[0040] In some cases, the steps of the above loop can be paused or terminated when one or more of the following conditions occur:

[0041] The loop execution can be paused or terminated when an edge computing node experiences power loss, restart, hardware failure, or overheating protection; it can also be paused or terminated when a ring main unit experiences a serious fault that triggers protection interlocking. These serious faults can include short-circuit faults, arcing faults, gas leak faults, and over-temperature faults. The loop execution can also be paused or terminated when the local digital twin model fails to perform continuous verification during self-evolution, experiences severe model drift, or is damaged. Furthermore, the loop execution can be paused or terminated when the entire system enters maintenance mode, receives a remote stop command from the cloud platform or local maintenance terminal, or is undergoing firmware upgrades; and it can be paused or terminated when all core sensors fail, making it impossible to obtain valid multimodal raw data.

[0042] Specifically, when only cloud communication is interrupted while all other units are functioning normally, the loop execution can continue without stopping. In this case, the edge computing node can automatically switch to the edge local independent closed-loop operation mode and continue executing steps S101 to S104 and S106. After cloud communication is restored, the edge computing node can re-execute step S105 to restore edge-cloud collaboration.

[0043] Optionally, in some embodiments, obtaining a standardized multimodal feature set based on the multimodal raw data during the operation of the ring main unit, i.e., step S101, may include:

[0044] Step S1011: Obtain the multimodal raw data; the multimodal raw data includes at least one of the following: electrical quantities, environmental quantities, mechanical state quantities, and insulation state quantities;

[0045] Step S1012: Perform signal denoising, outlier removal, missing value interpolation and completion, and dimension normalization on the multimodal raw data.

[0046] Step S1013: Timestamps are labeled on the processed multimodal raw data using a unified time reference source, and feature vectors under the same timestamps are retrieved using a sliding window to complete time alignment.

[0047] Step S1014: Based on the physical installation location coordinates of the sensor, perform spatial mapping on the aligned feature vectors to complete spatial coordinate alignment, and output the standardized multimodal feature set.

[0048] In step S1011, the edge computing node can acquire the multimodal raw data generated during the operation of the ring main unit. Combined with... Figure 3 As shown, the aforementioned multimodal raw data can include the following four categories of information: The first category is electrical quantities (corresponding to the electrical quantity data in the figure). Electrical quantities can include three-phase voltage, three-phase current, active power, reactive power, and frequency. The second category is environmental quantities (corresponding to the environmental quantity data in the figure). Environmental quantities can include the temperature, humidity, sulfur hexafluoride (SF6) gas density, SF6 gas pressure, and condensation state inside the ring main unit. The third category is mechanical state quantities (corresponding to the mechanical state data in the figure). Mechanical state quantities can include the circuit breaker opening and closing coil current, energy storage motor status, number of operations, and mechanical position. The fourth category is insulation state quantities (corresponding to the insulation state data in the figure). Insulation state quantities can include partial discharge UHF signals, partial discharge ultrasonic signals, and vibration signals. By collecting the above-mentioned multi-dimensional data, including electrical quantities, environmental quantities, mechanical state quantities, and insulation state quantities, the edge computing node can achieve full-state perception of the ring main unit.

[0049] In step S1012, the edge computing node can use the preprocessing unit to preprocess the acquired multimodal raw data. This preprocessing process may include signal denoising, outlier removal, missing value interpolation and completion, and dimension normalization.

[0050] Specifically, edge computing nodes can use filtering algorithms to denoise the acquired multimodal raw signals to eliminate high-frequency interference components. Edge computing nodes can also remove abnormal values ​​that are significantly beyond the sensor's measurement range or physically impossible. For missing data caused by communication packet loss or momentary sensor failure, edge computing nodes can use linear interpolation or nearest-neighbor interpolation algorithms to complete the data and normalize data with different physical dimensions to the same numerical range. For example, a current value of 800A and a temperature value of 85℃ can be normalized to a numerical range of 0 to 1. Through these preprocessing operations, edge computing nodes can obtain cleaned and standardized multimodal data.

[0051] In step S1013, the edge computing node can utilize a unified time alignment mechanism, employing the Global Positioning System (GPS) or the BeiDou Navigation Satellite System as a unified time reference source, to annotate each preprocessed frame of data with millisecond-level timestamp precision. The edge computing node can use a fixed-length sliding window, such as a 1-second window, to retrieve all feature vectors whose timestamps fall between the window's start and end times, combining these feature vectors into a data set representing the same time segment, thereby completing the time alignment between data from different sensors.

[0052] In step S1014, the edge computing node can map the corresponding feature values ​​in the aligned feature vectors to the corresponding spatial point coordinates in the digital twin model based on the physical installation coordinates of each sensor within the ring main unit through a multimodal feature fusion process. For example, if the installation coordinates of the temperature sensor at the A-phase cable connector are (X1, Y1, Z1), the edge computing node can map the feature value corresponding to the A-phase temperature in the aligned feature vectors to the spatial point with coordinates (X1, Y1, Z1) in the digital twin model. Following this analogy, the edge computing node can complete the spatial coordinate alignment of all feature vectors and output a standardized multimodal feature set.

[0053] It is understood that the time alignment and spatial alignment process described above in this embodiment adopts a unified high-precision timestamp combined with a sliding window time-series correlation method. By using timestamps and spatial coordinates, time synchronization and spatial synchronization between multi-rate and multi-source data are achieved, thereby realizing the fusion of multimodal features.

[0054] Optionally, in some embodiments, the step of performing a virtual-real consistency comparison between the standardized multimodal feature set and the output data of the local digital twin model to obtain state deviations, and synthesizing a global deviation based on each of the state deviations, i.e., step S102 may include:

[0055] Step S1021: Compare the standardized multimodal feature set with the output value of the local digital twin model to obtain a state deviation set that includes at least one or more of the following: temperature deviation, current deviation, partial discharge signal deviation, mechanical action time deviation, and insulation parameter deviation.

[0056] Step S1022: According to the preset on-site working condition calibration weights, the various state deviations in the state deviation set are weighted and summed to obtain the global deviation.

[0057] In step S1021, the edge computing node can compare the measured data contained in the standardized multimodal feature set with the predicted data output by the local digital twin model at the same time coordinate and spatial location coordinate. The edge computing node can calculate the temperature deviation, current deviation, partial discharge signal deviation, mechanical action time deviation, and insulation parameter deviation respectively, and summarize the above different types of deviations to form a state deviation set.

[0058] As an example, regarding temperature deviation, the measured temperature at the A-phase cable joint is 85℃, while the predicted temperature output by the local digital twin model at that location is 82℃, a difference of +3℃. Regarding current deviation, the measured current of phase A is 800A, while the predicted current output by the local digital twin model is 780A, a difference of +20A. Regarding mechanical action time deviation, the measured tripping time of the circuit breaker is 35ms, while the predicted tripping time output by the local digital twin model is 32ms, a difference of +3ms. Using the same method, the edge computing node can continue to calculate the partial discharge signal deviation and insulation parameter deviation. Through the above comparison operations, the edge computing node can obtain a set of state deviations containing various deviation types.

[0059] In step S1022, the edge computing node can call a set of weighting coefficients pre-calibrated for the current operating conditions of the ring main unit. For example, under peak power consumption conditions, the edge computing node can assign higher weights to current deviation and temperature deviation. Subsequently, the edge computing node can multiply each deviation in the state deviation set by its corresponding weighting coefficient, and sum the weighted deviations to calculate the global deviation value.

[0060] Optionally, in some embodiments, the step of determining the deviation source features and corresponding twin model nodes based on the state deviation, i.e., step S103, may include:

[0061] Step S1031: Determine the source characteristics of the deviation based on the correlation between the state deviation and the device topology.

[0062] Step S1032: Determine the corresponding twin sub-model node based on the mapping relationship of the deviation source features in the digital twin model.

[0063] In step S1031, the edge computing node can perform source tracing analysis based on the correlation between various state deviations in the ring main unit equipment topology. For example, when multiple deviations related to a specific component occur simultaneously, the edge computing node can determine that these deviations have a common source, thereby identifying the physical parameters or performance indicators that cause these deviations as deviation source characteristics. The A-phase current deviation, A-phase temperature deviation, and A-phase partial discharge deviation are all topologically associated with the A-phase cable joint node. The edge computing node can therefore determine that these three deviations have a common source, thus identifying the deviation source characteristic as the contact resistance or insulation status parameter of the A-phase cable joint. Alternatively, the system can further differentiate based on the spectral characteristics of the partial discharge signal. If the partial discharge signal is dominated by ultra-high frequency components, the deviation source characteristic can be identified as insulation aging parameters.

[0064] In step S1032, the edge computing node can locate itself based on the predetermined mapping relationship between the determined deviation source features and the digital twin model. The digital twin model can include a geometric model, an electrical model, a thermodynamic model, a mechanical model, and an insulation model, forming a multiphysics coupling model. The edge computing node can find the parameter node or functional module corresponding to the deviation source feature in the above multiphysics coupling model. This parameter node or functional module is the twin sub-model node, so that subsequent correction operations can be performed on this node.

[0065] Specifically, after acquiring the various state deviations, the system does not blindly update the parameters of the entire model. Instead, it analyzes the magnitude of individual state deviations to pinpoint specific source characteristics of the deviations. For example, if the partial discharge signal deviation is too large, the system can identify an insulation state anomaly. Based on this identification, the system can accurately locate the corresponding twin model node, such as the insulation state parameter node, so that targeted correction operations can be performed on that node subsequently.

[0066] As another example, when the deviation source characteristic is the contact resistance of the phase A cable joint, this deviation source characteristic is mapped to the contact resistance parameter node under the electrical sub-model in the digital twin model. The edge computing node then determines the corresponding twin sub-model node as the contact resistance parameter node so that a correction operation can be performed on the contact resistance parameter node subsequently.

[0067] Optionally, in some embodiments, triggering the hierarchical self-evolutionary correction of the local digital twin model based on the comparison relationship between the global deviation and the preset threshold, i.e., step S104 may include:

[0068] When the global deviation is less than or equal to the first threshold, it is determined to be a normal operating condition. The correction increment of the basic physical parameters is calculated by the lightweight gradient descent algorithm, and the parameters of the twin model nodes corresponding to the deviation source features are fine-tuned.

[0069] When the global deviation is greater than the first threshold and less than or equal to the second threshold, it is determined to be an abnormal warning condition. Abnormal modal features are extracted and used as training labels. The local network corresponding to the deviation source features is incrementally trained on the edge side, and the updated local network weights are transferred to the local digital twin sub-model through the model distillation method.

[0070] When the global deviation is greater than the second threshold, it is determined to be a fault-locked-off condition. The write entry of the local digital twin sub-model is blocked to force the parameter update to freeze, and the local hardware acceleration channel is enabled to perform fault logic reasoning.

[0071] Specifically, when the global deviation is less than or equal to the first threshold, the edge computing node can determine that the ring main unit is in normal operating condition. At this time, the edge computing node can calculate the correction increment of the basic physical parameters using a lightweight gradient descent algorithm, and fine-tune the parameters of the twin model node corresponding to the deviation source feature, thereby gradually approximating the output of the local digital twin model to the measured value of the physical entity. For example, for the contact resistance parameter of phase A, the edge computing node can calculate a correction increment of 0.01 milliohms per iteration using the lightweight gradient descent algorithm, and perform parameter fine-tuning on the twin model node corresponding to the deviation source feature, i.e., the phase A contact resistance parameter node, so that the output of this node gradually approaches the measured value.

[0072] In some examples, the fundamental physical parameters include at least one of the following: resistivity, heat transfer coefficient, and coefficient of mechanical friction.

[0073] When the global deviation is greater than the first threshold and less than or equal to the second threshold, the edge computing node can determine that the ring main unit is in an abnormal early warning condition. At this time, the edge computing node can extract abnormal mode features, such as specific harmonic components of the A-phase current and abnormal temperature rise rates. Using these extracted abnormal mode features as training labels, incremental training is performed on the local network corresponding to the deviation source features at the edge. This local network can be, for example, an A-phase cable connector sub-model. After completing the incremental training, the edge computing node can transfer the updated local network weights to the local digital twin sub-model using a model distillation method.

[0074] When the global deviation exceeds the second threshold, the edge computing node can determine that the ring main unit is in a fault-locked state. At this time, the edge computing node can immediately block the parameter write entry of the local digital twin sub-model to forcibly freeze all parameter updates, thereby preventing uncontrollable drift of the model. At the same time, the edge computing node can enable local hardware acceleration channels, such as field-programmable gate arrays (FPGAs) or graphics processing units (GPUs), to perform fault logic reasoning and output fault alarm information.

[0075] In some examples, the first threshold and the second threshold can be pre-calibrated based on the rated parameters and historical operating data of the ring main unit. For example, the first threshold can be set to 0.05 to 0.1, and the second threshold can be set to 0.3 to 0.5. The above thresholds can also be dynamically adjusted according to the on-site operating conditions, for example, when seasonal load changes occur, the updated threshold parameters can be issued by the cloud platform. This embodiment does not specifically limit this.

[0076] It should be noted that the above steps involve the core self-evolution mechanism of the digital twin model. Combining the electrical, thermodynamic, mechanical, and insulation models of the ring main unit to form a multiphysics model, the system can employ adaptive threshold triggering logic. Specifically, if the global state deviation vector is less than or equal to the threshold, it indicates that the current model is valid, and the system does not need to trigger the evolution operation, thus saving edge-side computing resources; if the global state deviation vector is greater than the threshold, the system triggers a self-evolution correction operation.

[0077] Unlike related technologies that use models such as Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), or Transformers for full retraining in the cloud, this embodiment specifically adopts a lightweight edge-specific neural network architecture based on physical mechanism constraints to adapt to the limited computing resources of edge computing nodes.

[0078] Specifically, the lightweight edge-specific neural network can be logically divided into a common feature extraction subnetwork and a physical attribute decoupling subnetwork. The common feature extraction subnetwork can be used to process the temporal correlation information in the multimodal raw data; the physical attribute decoupling subnetwork can contain multiple independent parameter branches, each branch corresponding to a physical subsystem of the ring main unit, such as a contact resistance branch, an insulation medium branch, and a mechanism friction branch.

[0079] The system can construct lightweight edge-specific neural networks by performing structured pruning and quantization compression operations on pre-trained digital twin models. Specifically, the system can remove redundant neuron connections from the model and perform low-bit quantization on the retained weight parameters, such as converting floating-point parameters to 8-bit integer parameters. Through these operations, the system can compress the overall model parameter count by more than 80% while maintaining the core topology and input-output mapping of the physical model.

[0080] When performing hierarchical self-evolutionary correction, edge computing nodes can adopt targeted neuron activation update strategies. Specifically, edge computing nodes can retrieve and lock the weights of common feature extraction layers that are unrelated to the twin model node located in step S103, keeping them stable; at the same time, edge computing nodes limit the computational scope of the incremental gradient descent algorithm to specific branch neurons in the physical property decoupling layer that are directly related to the deviation source features.

[0081] As an example, if the source of deviation is contact resistance, the edge computing node can update only the local parameter layer in the electro-thermal mapping branch without triggering the calculation of the insulation or mechanical branches. This local parameter update mechanism avoids a full reconstruction and retraining of the entire digital twin model. In practical edge computing nodes, the combined time consumption of a single online inference and parameter correction can be controlled within 100 milliseconds, thus meeting the response speed requirements in real-time monitoring scenarios for ring main units.

[0082] After completing local incremental training at the edge, edge computing nodes can introduce an offline knowledge distillation mechanism to further optimize model performance. Specifically, the cloud platform can use the trained global master model as the teacher model and periodically distribute the soft label probability distribution output by this teacher model to the edge. The edge computing nodes use the soft labels distributed from the cloud as supervision signals to perform distillation training on the corrected local sub-model, calculating the soft target loss between the local sub-model output and the soft labels, for example, using KL divergence as the loss function for backpropagation. In this way, edge computing nodes can transfer knowledge from the global master model to the local sub-model without deploying the complete teacher model at the edge. This mechanism helps to correct overfitting bias that may be introduced by insufficient samples at the edge and further reduces model redundancy.

[0083] Furthermore, edge computing nodes can incorporate historical state-based transfer learning strategies to enhance model generalization capabilities. Specifically, edge computing nodes can extract the multimodal feature distribution of the ring main unit under historical normal operating conditions, using this distribution as the source domain; simultaneously, they can extract the feature distribution under current abnormal operating conditions or during the evolution process, using this distribution as the target domain. Edge computing nodes can employ the maximum mean difference algorithm as a metric, projecting the source and target domains onto the regeneration kernel Hilbert space through a mapping function and minimizing the distribution distance between them, thereby performing domain-adaptive processing. Through this transfer learning strategy, even in environments with scarce sample data on the edge side, edge computing nodes can maintain high model generalization capabilities, utilizing prior knowledge from historical normal states to identify the equipment evolution trend under current operating conditions, thus avoiding overfitting due to insufficient local samples.

[0084] Furthermore, the training loss function of the lightweight edge-specific neural network may include a physical residual term. This physical residual term can be constructed based on Kirchhoff's laws or the thermal balance equation of a ring network cabinet, and is used to constrain the weight update direction of neurons, thereby ensuring that the evolved model parameters always remain within a physically reasonable numerical range.

[0085] Optionally, in some embodiments, after triggering the hierarchical self-evolution correction of the local digital twin model, the method further includes: determining whether the residual between the output value of the hierarchically self-evolution correction local digital twin model and the measured value of the physical entity meets a preset accuracy threshold; determining whether the corrected physical parameters are within the rated operating range of the ring main unit model; if the above two determinations are met, the verification passes and the new version of the local digital twin model is enabled; if the above two determinations are not met, a supplementary iteration is triggered; if the verification fails again, an automatic version rollback to the previous stable version is executed.

[0086] Specifically, after triggering the hierarchical self-evolutionary correction of the local digital twin model, the method further includes a model verification and version rollback process. Specifically, the edge computing node can determine whether the residual between the output value of the hierarchically self-evolutionarily corrected local digital twin model and the measured value of the physical entity meets a preset accuracy threshold (e.g., residual ≤ 1%). Simultaneously, the edge computing node can determine whether the corrected physical parameters are within the rated operating range of the ring main unit model. If both of the above judgments are satisfied, the edge computing node determines that the verification has passed and activates the new version of the local digital twin model. If the above two judgments are not satisfied, the edge computing node can trigger a supplementary iteration. If the verification fails again, the edge computing node can perform automatic version rollback, restoring the local digital twin model to the previous stable version to ensure the industrial-grade safety of the ring main unit operation.

[0087] Optionally, in some embodiments, the step of driving the edge-cloud communication unit to perform hierarchical adaptive synchronous transmission to the cloud platform based on the hierarchical self-evolution correction operating condition results, i.e., step S105 may include:

[0088] When in normal operating condition, low-frequency periodic synchronization mode is activated, and only incremental data packets with fine-tuned parameters are uploaded;

[0089] When the abnormal warning condition is underway, a high-frequency real-time push mechanism is triggered to upload the abnormal modal features and the updated local network weights.

[0090] When the fault-locked working condition is in effect, the communication interception strategy is activated to intercept requests for uploading regular monitoring data to the cloud and prioritize the transmission of fault alarm commands, storing the transient waveform data of the fault to the local circular buffer.

[0091] The embodiments of this application can adopt a three-level adaptive synchronization strategy to automatically switch the edge-cloud collaboration mechanism according to the different operating conditions of the ring network cabinet.

[0092] Specifically, when the edge computing node is in normal operating condition after hierarchical self-evolution correction, it can activate a low-frequency periodic synchronization mode to perform incremental lightweight synchronization. Under normal operating conditions, the edge computing node only uploads the incremental data packets generated by parameter fine-tuning correction. This incremental data packet can include the adjusted twin model node identifier and the corrected parameter values, without needing to upload the complete local digital twin model state. For example, the edge computing node can upload incremental data packets to the cloud platform every 5 to 10 minutes. This design can reduce the computing power consumption at the edge and the bandwidth pressure on the cloud platform.

[0093] When the edge computing node is in an anomaly warning state after hierarchical self-evolution correction, it can trigger a high-frequency real-time push mechanism (also known as full-volume high-precision synchronization mode). Triggering conditions for the anomaly warning state can include abnormal temperature, excessive partial discharge, or sudden current changes. In this mode, the synchronization period can be set to less than or equal to 100ms. The edge computing node can upload full multimodal data and complete local digital twin sub-model parameters. The uploaded content can include anomaly modal features and updated local network weight files. The anomaly modal features can include abnormal harmonic spectrum data and current-temperature variation curves. This data can be used by the cloud platform for in-depth analysis and long-term trend assessment, thereby ensuring the high accuracy of the digital twin model and facilitating fault location.

[0094] When the edge computing node is in a fault-locked state after hierarchical self-evolution correction, it can activate a communication interception strategy to perform independent local locking synchronization. Triggering conditions for the fault-locked state can include short-circuit faults, arcing faults, severe partial discharge, or gas leaks. Under this state, the edge computing node can stop uploading routine monitoring data to the cloud platform and instead perform independent computation processing locally. The edge computing node can immediately transmit fault alarm commands to the cloud platform with the highest priority, while simultaneously locking the local digital twin sub-model to prevent erroneous updates. The edge computing node can store detailed waveform data for a period of time before and after the fault occurrence in a local circular buffer, such as waveform data from 0.5 seconds before the fault to 1.5 seconds after the fault. After the fault alarm command is sent, the edge computing node can decide whether to retransmit the aforementioned waveform data based on instructions from the cloud platform. This prioritizes the accuracy of fault diagnosis without losing critical fault data.

[0095] Optionally, in some embodiments, after the driving edge-cloud communication unit performs hierarchical adaptive synchronous transmission to the cloud platform, the method may further include:

[0096] Step S201: Receive the optimization strategy generated by the cloud platform based on the hierarchical adaptive synchronous transmission data; the optimization strategy includes optimization suggestions for the temperature and humidity compensation coefficient and partial discharge early warning threshold of the environment where the ring network cabinet is located;

[0097] Step S202: According to the optimization strategy, update the configuration parameters of the local digital twin sub-model to achieve a global twin closed loop for edge-cloud collaboration.

[0098] For step S201, see, for example, [link to example]. Figure 4As shown, edge computing nodes can receive optimization strategies generated by the cloud platform (cloud side) through hierarchical adaptive synchronous data transmission. Specifically, the cloud platform can utilize long-term operational data accumulated in its big data storage to perform in-depth analysis and global optimization through a global digital twin master model. The generated optimization strategies can be distributed to the edge side via the strategy optimization and model distribution functions through edge-cloud communication units (edge-to-cloud, data-to-network communication). For example, based on the analysis results of long-term operational data, the cloud platform can suggest adjusting the temperature compensation coefficient from 0.95 to 0.97. As another example, to improve early warning sensitivity, the cloud platform can suggest adjusting the partial discharge early warning threshold for the ring main unit from 50 pC to 45 pC. The above optimization strategies can effectively compensate for the model generalization problem that may be caused by insufficient historical samples on the edge side.

[0099] In step S202, the edge computing node can update the configuration parameters of its local digital twin model based on the received optimization strategy, thereby leveraging the global computing power of the cloud platform to improve the accuracy of the local digital twin model. Specifically, combined with Figure 4 As shown, edge computing nodes can use incremental update mechanisms to modify settings such as temperature compensation coefficients and partial discharge warning thresholds in their local models, making the local synchronization effect more closely match actual operating conditions. Through these operations, real-time data acquired through multimodal sensing can be combined with global strategies issued from the cloud, achieving a global digital twin closed loop for edge-cloud collaboration. This not only enables the local digital twin sub-model to continuously improve its accuracy but also supports global visualization on the cloud side, ensuring a high degree of consistency between the physical and digital worlds throughout their entire lifecycle.

[0100] It is not difficult to see that this application has at least the following beneficial effects:

[0101] In related technologies, the operation of ring main units involves data from multiple types of sensors, including electrical, environmental, mechanical, and insulation sensors. Different sensors differ in sampling frequency, time reference, and installation location. Directly using this multimodal data for virtual-to-real comparison can easily lead to time misalignment and spatial mismatch, resulting in low synchronization accuracy. This application's embodiment obtains a standardized feature set by spatiotemporally aligning the original multimodal data, enabling the comparison of measured data and model prediction data at the same time and spatial coordinates. This reduces errors caused by asynchrony at the data source, helping to improve the accuracy of virtual-to-real consistency comparison.

[0102] In related technologies, edge computing nodes and cloud platforms often use fixed-frequency or full-data-upload methods for data interaction. Regardless of the operating conditions of the ring network cabinet, it consumes similar bandwidth resources, which can easily lead to bandwidth waste during normal operation and failure to send critical data in a timely manner during anomalies, resulting in significant pressure on edge-cloud interaction. This application's embodiment performs hierarchical adaptive synchronous transmission based on the hierarchical self-evolutionary correction of the operating conditions. During normal operation, only incremental data packets are uploaded to reduce bandwidth consumption. During anomaly warnings, the upload frequency is increased and key features and model weights are pushed. During fault blocking, alarm commands are prioritized for transmission while regular data is intercepted. This dynamic adjustment of the synchronization strategy based on actual operating conditions helps alleviate the transmission pressure on the edge-cloud communication link.

[0103] In related technologies, the update mechanism of digital twin models often lacks differentiation of operating conditions. This leads to either frequent updates wasting computational resources or delayed updates causing excessive deviations between the model and the physical entity, making it difficult to adaptively adjust and correct behavior under different operating conditions such as normal, warning, and fault. The embodiments of this application trigger hierarchical self-evolutionary correction based on the comparison between the global deviation and a preset threshold. When the deviation is small, only parameter fine-tuning is performed to save computational power; when the deviation is moderate, local incremental training is performed to improve model accuracy; and when the deviation is too large, model updates are frozen and fault inference is enabled to ensure safety. This allows for the selection of appropriate correction methods under different operating conditions, helping to improve the system's adaptability to changes in operating conditions.

[0104] In summary, the embodiments of this application improve data synchronization accuracy through spatiotemporal alignment, alleviate edge-cloud interaction pressure through hierarchical adaptive synchronization, and enhance operating condition adaptability through hierarchical self-evolution correction. These measures can, to a certain extent, solve the technical problems of low virtual-real synchronization accuracy, high edge-cloud interaction pressure, and poor operating condition adaptability of multimodal data in ring network cabinets in related technologies.

[0105] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but without changing the core design of the algorithm and process, are also within the scope of protection of this application.

[0106] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent from this embodiment.

[0107] Furthermore, some embodiments of this application also provide a terminal device, specifically a ring main unit intelligent edge terminal. The terminal device can be various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframes, etc. The terminal device can also be various forms of mobile devices, such as cellular phones, smartphones, wearable devices, and other similar computing devices.

[0108] The terminal device includes: one or more processors; and a memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the methods provided in any one or more of the above embodiments. Figure 5 An exemplary structural diagram of the terminal device is disclosed. The terminal device includes one or more processors 1101, a memory 1102, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the terminal device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple terminal devices can be connected, each providing some of the necessary operations. The components, their connections and relationships, and their functions shown herein are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0109] The terminal device may further include an input device 1103 and an output device 1104. The processor 1101, memory 1102, input device 1103 and output device 1104 may be connected by a bus or other means, as shown in the figure, which is connected by a bus.

[0110] Input device 1103 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the terminal device, such as a touch screen, keypad, mouse, trackpad, touchpad, pointer, one or more mouse buttons, trackball, joystick, etc. Output device 1104 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display, a light-emitting diode display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0111] To provide interaction with the user, the terminal device can be a computer. The computer has: a display device (e.g., a cathode ray tube or LCD monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback); and input from the user can be received in any form (e.g., voice input or tactile input).

[0112] In this embodiment, a computer-readable medium stores a computer program / instructions, which, when executed by a processor, implement the steps of the methods provided in any one or more of the above embodiments. This computer-readable medium may be included in the terminal device described in the above embodiments; or it may exist independently and not assembled into that device. The aforementioned computer-readable medium carries one or more computer-readable instructions.

[0113] The memory 1102 can serve as a non-transitory computer-readable storage medium, used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 1102, thereby implementing the program instructions / modules corresponding to the methods provided in any one or more of the embodiments described above in this application.

[0114] The memory 1102 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 1102 may include high-speed random access memory and may also include non-transient memory, such as at least one disk storage device, flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 1102 may optionally include memory remotely located relative to the processor 1101, and these remote memories can be connected to the terminal device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0115] It should be noted that the computer-readable medium described in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. Computer-readable media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0116] Computer-readable media include permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, read-only optical discs, digital versatile optical discs or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0117] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including local area networks (LANs) or wide area networks (WANs), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0118] In the above embodiments, all or part of the implementation can be achieved through software, hardware, firmware, or any combination thereof. For example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of this application can be executed by a processor to implement the above steps or functions. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, magnetic or optical drives, floppy disks, and similar devices. In addition, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0119] The computer program product provided in this application includes one or more computer programs / instructions. When executed by a processor, these computer programs / instructions generate, in whole or in part, the processes or functions described in this application. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0120] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0121] The scope of this application is defined by the appended claims rather than the foregoing description, and is therefore intended to encompass all variations falling within the meaning and scope of equivalents of the claims. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural. Terms such as "first," "second," etc., are used only to distinguish descriptions and do not indicate any particular order, nor should they be construed as indicating or implying relative importance.

[0122] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily made by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims, and the above embodiments should be regarded as exemplary and non-limiting.

Claims

1. A method for self-evolution and synchronization of digital twins in ring main units, characterized in that, The method is applied to edge computing nodes, and the method includes: Based on the multimodal raw data during the operation of the ring main unit, a standardized multimodal feature set is obtained; the standardized multimodal feature set includes feature values ​​aligned with timestamps and spatial coordinates; The standardized multimodal feature set and the output data of the local digital twin model are compared for virtual-real consistency to obtain state deviations, and a global deviation is synthesized based on the state deviations of each item. Based on the state deviation, determine the deviation source features and the corresponding twin model nodes; Based on the comparison relationship between the global deviation and the preset threshold, the local digital twin sub-model is triggered to perform hierarchical self-evolution correction on the twin sub-model node corresponding to the deviation source feature; Based on the operating conditions of the hierarchical self-evolution correction, the edge-cloud communication unit is driven to perform hierarchical adaptive synchronous transmission to the cloud platform; Repeat the above steps until a stop command is received or the protection shutdown condition is triggered.

2. The method according to claim 1, characterized in that, The standardized multimodal feature set obtained based on the multimodal raw data during the operation of the ring main unit includes: Obtain the multimodal raw data; the multimodal raw data includes at least one of the following: electrical quantities, environmental quantities, mechanical state quantities, and insulation state quantities; The original multimodal data is subjected to signal denoising, outlier removal, missing value interpolation and completion, and dimension normalization. The processed multimodal raw data is timestamped using a unified time reference source, and feature vectors with the same timestamp are retrieved using a sliding window to complete time alignment. Based on the physical installation location coordinates of the sensor, the aligned feature vectors are spatially mapped to complete spatial coordinate alignment, and the standardized multimodal feature set is output.

3. The method according to claim 1, characterized in that, The step of comparing the standardized multimodal feature set and the output data of the local digital twin model to obtain state deviations, and synthesizing a global deviation based on each of the state deviations, includes: The standardized multimodal feature set and the output value of the local digital twin model are compared to obtain a state deviation set that includes at least one or more of the following: temperature deviation, current deviation, partial discharge signal deviation, mechanical action time deviation, and insulation parameter deviation. According to the preset on-site working condition calibration weights, the various state deviations in the state deviation set are weighted and summed to obtain the global deviation; The step of determining the deviation source features and corresponding twin sub-model nodes based on the state deviation includes: determining the deviation source features based on the correlation of the state deviation in the device topology; and determining the corresponding twin sub-model nodes based on the mapping relationship of the deviation source features in the digital twin model.

4. The method according to claim 1, characterized in that, The step of triggering the hierarchical self-evolutionary correction of the local digital twin model based on the comparison relationship between the global deviation and the preset threshold includes: When the global deviation is less than or equal to the first threshold, it is determined to be a normal operating condition. The correction increment of the basic physical parameters is calculated by the lightweight gradient descent algorithm, and the parameters of the twin model nodes corresponding to the deviation source features are fine-tuned. When the global deviation is greater than the first threshold and less than or equal to the second threshold, it is determined to be an abnormal warning condition. Abnormal modal features are extracted and used as training labels. The local network corresponding to the deviation source features is incrementally trained on the edge side, and the updated local network weights are transferred to the local digital twin sub-model through the model distillation method. When the global deviation is greater than the second threshold, it is determined to be a fault-locked-off condition. The write entry of the local digital twin sub-model is blocked to force the parameter update to freeze, and the local hardware acceleration channel is enabled to perform fault logic reasoning.

5. The method according to claim 4, characterized in that, After triggering the hierarchical self-evolutionary correction of the local digital twin model, the method further includes: Determine whether the residual between the output value of the local digital twin model after the hierarchical self-evolution correction and the measured value of the physical entity meets the preset accuracy threshold. Determine whether the corrected physical parameters are within the rated operating range of the ring main unit model; If both of the above conditions are met, the verification passes and the new version of the local digital twin model is enabled; If the above two conditions are not met, a supplementary iteration is triggered. If the verification fails again, an automatic version rollback to the previous stable version is executed.

6. The method according to claim 4, characterized in that, The step of driving the edge-cloud communication unit to perform hierarchical adaptive synchronous transmission to the cloud platform based on the hierarchical self-evolution correction working condition results includes: When in normal operating condition, low-frequency periodic synchronization mode is activated, and only incremental data packets with fine-tuned parameters are uploaded; When the abnormal warning condition is underway, a high-frequency real-time push mechanism is triggered to upload the abnormal modal features and the updated local network weights. When the fault-locked working condition is in effect, the communication interception strategy is activated to intercept requests for uploading regular monitoring data to the cloud and prioritize the transmission of fault alarm commands, storing the transient waveform data of the fault to the local circular buffer.

7. The method according to claim 1, characterized in that, After the driving edge-cloud communication unit performs hierarchical adaptive synchronous transmission to the cloud platform, the method further includes: The system receives an optimization strategy generated by the cloud platform based on the hierarchical adaptive synchronous transmission data. The optimization strategy includes optimization suggestions for the temperature and humidity compensation coefficient and partial discharge early warning threshold of the environment in which the ring main unit is located. Based on the optimization strategy, the configuration parameters of the local digital twin sub-model are updated to achieve a global twin closed loop for edge-cloud collaboration.

8. A terminal device, characterized in that, The terminal device includes: One or more processors; and A memory storing computer program instructions, which, when executed, cause the processor to perform the steps of the method as described in any one of claims 1 to 7.

9. A computer-readable medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 7.

Images