Internet of things based intelligent box-type substation cluster management operation system

Abstract

The application relates to the technical field of power system automation and artificial intelligence, and discloses an intelligent box-type transformer substation cluster management operation system based on Internet of Things, which adopts a three-layer collaborative architecture of end-edge-cloud, comprises a sensing and executing module, an edge computing gateway and a cloud cluster management platform, the edge computing gateway decomposes total feature deviation along temperature, aging and load dimensions through drift detection, distinguishes between normal evolution and faults of equipment, triggers model updating for normal drift, the cloud updates model parameters through cluster correlation analysis and incremental learning, realizes self-adaptive evolution of a diagnosis model according to equipment states, and a cloud cluster health balance scheduling engine takes the sum of equivalent life loss increments weighted by traditional power targets and differentiated aging costs as an objective function, applies differentiated load constraints to box-type transformers in different aging stages, and realizes scheduling transformation of balanced life loss.

Description

Technical Field

[0001] This invention relates to the field of power system automation and artificial intelligence, and in particular to an intelligent prefabricated substation cluster management operating system based on the Internet of Things. Background Technology

[0002] With the in-depth advancement of the construction of new power systems, distributed photovoltaic, energy storage, and electric vehicle charging facilities are being connected to the distribution network on a large scale. As a key node at the end of the distribution network, prefabricated substations have generally reached a single unit capacity of 500-2000kVA, and the scale of clusters often ranges from dozens to hundreds of units. This places unprecedentedly stringent requirements on the reliability of equipment operation, the speed of fault self-healing, and the overall service life of the cluster. Due to their advantages such as strong real-time perception and high efficiency of local data processing, IoT sensor networks and edge computing gateways have become the mainstream solutions for realizing intelligent operation and maintenance of prefabricated substations. As the core control center of the prefabricated substation cluster, the cloud-based collaborative operating system undertakes the key functions of equipment status monitoring, fault diagnosis and early warning, power supply self-healing scheduling, and full life cycle management.

[0003] For example, Chinese patent CN114498934A discloses a substation monitoring system. The system includes multiple data acquisition devices, an edge computing platform set up on the field side, and a cloud platform. The cloud platform uses historical monitoring data uploaded by the edge computing platform to perform comprehensive learning and training, obtains a big data model, and sends it to the edge computing platform. The edge computing platform performs anomaly analysis on the current monitoring data based on the big data model and uploads the anomaly results to the auxiliary equipment centralized monitoring system and the cloud platform.

[0004] The above-mentioned and existing related technologies often have the following drawbacks:

[0005] 1. Fault diagnosis models based on historical monitoring data and fixed deployments are ill-suited to adapt to the characteristic shifts in the operating status of prefabricated substations over time. Prefabricated substations are long-term operating power equipment, and the monitoring characteristics of their components, such as cores, windings, insulating oil, and heat dissipation structures, continuously change under different seasons, load levels, and aging stages, including vibration spectra, temperature distribution, and gas composition in the oil. Existing technologies typically use historical monitoring data to train diagnostic models and then distribute these models to the field for anomaly analysis. However, if the deployed models lack a continuous correction mechanism corresponding to the actual aging state of the equipment, seasonal environmental changes, and changes in operating conditions, mismatches between the diagnostic benchmark and the current state of the equipment can easily occur. For example, changes in insulating oil viscosity at low temperatures may cause fluctuations in vibration or temperature rise characteristics, while changes in heat dissipation conditions during high-temperature seasons may mask abnormal trends caused by early insulation aging, leading to false alarms, missed alarms, or decreased reliability of diagnostic results, thus affecting the long-term stable use of the intelligent operation and maintenance system for prefabricated substations.

[0006] 2. The current cluster operation and scheduling system lacks consideration for the differences in aging levels and lifespan losses among different prefabricated substations. This can easily lead to some equipment continuously bearing high operating pressure. The distributed photovoltaic, energy storage, electric vehicle charging facilities, and surrounding power loads connected to different prefabricated substations vary, resulting in differences in the long-term temperature rise levels, load surges, reverse power flows, and heat dissipation conditions experienced by the equipment. Consequently, the aging processes of the internal insulation materials, conductive connections, and heat dissipation components of each prefabricated substation are not consistent. Existing scheduling methods focus more on power balance, load distribution, and power supply continuity, but do not adequately consider the actual aging level, remaining carrying capacity, and lifespan loss trends of individual prefabricated substations. This may cause equipment already in the accelerated aging stage to continue bearing high loads, while equipment in better condition fails to share the operating pressure reasonably. After long-term operation, some equipment within the cluster may prematurely experience insulation degradation, abnormal temperature rise, or connection failures, thereby increasing the risk of centralized maintenance, outages, and decreased power supply reliability. Summary of the Invention

[0007] The technical problem to be solved by this invention is that the existing technology has the following drawbacks: the diagnostic model of the prefabricated substation cluster management has the following shortcomings: the accuracy of the diagnostic model has the following drawbacks: it has the following shortcomings: the cluster scheduling only focuses on power balance and ignores the different aging states of each prefabricated substation, resulting in concentrated loss of life. To this end, we propose an intelligent prefabricated substation cluster management operating system based on the Internet of Things.

[0008] To achieve the above objectives, this application adopts the following technical solution: an intelligent prefabricated substation cluster management operating system based on the Internet of Things, comprising:

[0009] The sensing and execution module deployed inside each prefabricated substation includes sensors for collecting equipment operating status parameters and actuators for performing opening and closing operations.

[0010] The edge computing gateway, installed on the field side of each prefabricated substation, includes a local diagnostic inference engine, a drift detection and identification unit, and a model update request unit. The local diagnostic inference engine loads an individual diagnostic model, performs online inference on the feature vectors extracted from multi-source data collected by sensors, and outputs anomaly scores. The drift detection and identification unit continuously monitors the statistical deviation of the distribution of the feature vectors relative to the baseline distribution at the initial stage of operation. When conceptual drift is detected, the total feature offset is decomposed into temperature-affected components, aging-affected components, and load change components to identify the dominant drift factors. For drifts determined to be caused by the normal evolution of the equipment's physical state, when the drift state tends to stabilize and the anomaly score output by the local diagnostic inference engine for the normal sample under the new normal continues to be lower than the preset alarm threshold, the model update request unit uploads the drift feature vectors to the cloud.

[0011] The cloud-based cluster management platform deployed in a remote center includes a model evolution unit and a cluster health and balancing scheduling engine. The model evolution unit receives the drift feature vector, generates updated individual model parameters through cluster-level correlation analysis and incremental learning, and distributes them to the corresponding edge computing gateway. The cluster health and balancing scheduling engine includes a bi-objective optimization solution submodule, which generates a scheduling scheme by optimizing the objective function during fault recovery and load distribution.

[0012] The optimization objective function is:

[0013]

[0014] in, For traditional power scheduling objectives, For the first Tabletop transformer in solution The equivalent aging increment generated below For the first The aging cost weight of each transformer substation is dynamically assigned by the cloud based on the current life stage label of each substation. Substations in the healthy or prime stage are given a lower weight to bear more load, while substations in the aging or accelerated aging stage are given a higher weight to be subject to stricter load increment constraints. and The trade-off coefficient between the two sub-objectives satisfies The normalization constraint.

[0015] Preferably, the sensors include sensors for measuring the temperature of winding hot spots, sensors for acquiring vibration spectra, sensors for online monitoring of dissolved gas components in oil, and sensors for capturing partial discharge signals; the actuators include electric operating mechanisms for incoming and outgoing line circuit breakers and electric operating mechanisms for tie switches; the edge computing gateway also includes a data aggregation and feature extraction unit, which receives multi-source data output from the sensors, performs time alignment and feature extraction, and converts the raw sampled data into structured feature vectors before inputting them into the local diagnostic inference engine.

[0016] Preferably, the drift detection and identification unit calculates the drift statistical distance based on the Mahalanobis distance between the mean of features within the sliding window and the mean of the baseline features in the initial stage of operation. The preset threshold is determined based on the 95th percentile of the statistical distribution of the drift statistical distance under normal operating conditions in the initial stage of operation. When the drift statistical distance continuously exceeds the preset threshold, it is determined that conceptual drift has occurred and the drift type identification process is triggered, rather than directly determining it as a device failure.

[0017] Preferably, the drift detection and identification unit decomposes the total characteristic offset into drift components caused by temperature environment, drift components caused by equipment aging, drift components caused by load change, and residual terms of sensor noise and unmodeled factors. By monitoring the time evolution of each component, the dominant drift factor is identified to determine whether the drift belongs to the normal evolution of the equipment's physical state. Among them, the drift component caused by temperature environment is related to the winding hot spot temperature and the ambient temperature and exhibits seasonal fluctuations; the drift component caused by equipment aging increases monotonically with the cumulative equivalent aging amount, both of which correspond to the normal evolution of the equipment's physical state; the drift component caused by load change is mapped to the load rate and power factor and exhibits short-term jumps during periods of impact load; when the drift component caused by temperature environment or the drift component caused by equipment aging dominates the total deviation and the drift component caused by load change is not significantly abnormal, it is determined to be normal drift and no fault alarm is triggered; when the residual terms after decomposition continue to increase and cannot be explained by known environmental or aging factors, it is marked as a potential fault signal and enters the fault diagnosis process.

[0018] Preferably, the cloud-based cluster management platform also includes a cluster health profiling unit, which collects structured health data uploaded by the edge computing gateways of each transformer substation and generates a cluster health profile containing life stage labels for each transformer substation. After receiving the drift feature vector, the model evolution unit retrieves the drift feature vectors of other transformer substations of the same model, batch, and operating under similar environmental conditions and compares them using cosine similarity as a consistency measure. If the cosine similarity between the drift feature vectors of multiple devices under similar environmental conditions is higher than a preset threshold, it is confirmed as a common environmental drift, and domain adaptive training is performed on the population base model. If only individual devices show a shift and it is strongly correlated with the cumulative equivalent aging amount, it is confirmed as an individual aging drift. Incremental learning is performed on the base model using an elastic weight consolidation method, and a secondary constraint term for the network parameters related to historical fault modes is introduced into the loss function to generate the individual model parameters of the transformer substation, so as to adapt the model to the new data distribution without forgetting the historical fault modes.

[0019] Preferably, the life stage label is dynamically refreshed by the cloud based on the health index and cumulative equivalent aging amount periodically uploaded by the edge computing gateway of each transformer. Transformers that are in the early stage of operation and have a low cumulative equivalent aging amount are marked as healthy period. Transformers with a cumulative equivalent aging amount reaching a certain proportion of the design life are marked as mature period. Transformers that are close to the critical value of insulation life are marked as aging period. Transformers that have detected latent defects or whose cumulative aging amount exceeds the warning threshold are marked as accelerated aging period.

[0020] Preferably, the edge computing gateway also includes an aging quantification unit, which calculates the aging acceleration factor based on the insulation thermal aging law using the winding hot spot temperature, accumulates the equivalent aging amount in hourly statistical windows, and periodically uploads the converted health index to the cloud cluster management platform, so that the cloud can directly obtain structured health parameters without receiving the original temperature sequence of high-frequency sampling; the equivalent aging increment in the objective function is calculated based on the accumulated equivalent aging amount.

[0021] Preferably, the cluster health balancing scheduling engine also includes a constraint application submodule. The constraints applied during the optimization solution process include: the load rate of each transformer does not exceed the upper limit of its current health status, which is estimated and uploaded in real time by the edge computing gateway based on the winding hot spot temperature margin; the load transfer path ensures that there is no islanding and no reverse power flow exceeding the limit; and the number of times the tie switch operates does not exceed the mechanical life limit of the switch within a preset period.

[0022] Preferably, after receiving the scheduling command from the cloud cluster management platform, the edge computing gateway first performs a local security condition check; reads the current health index and remaining life estimate of the transformer; if the load increment required by the command exceeds the preset life protection threshold of the device (the life protection threshold is the maximum load increment corresponding to the upper limit of the allowed winding hot spot temperature under the current health state of the transformer), it sends a rejection reason to the cloud and requests a regeneration of the scheduling scheme; for the tie switch closing command, it uses the real-time voltage amplitude, frequency and phase difference of the two busbars provided by the synchronization phasor measurement module set at the tie switch to perform synchronization condition detection; if the condition is not met, it is delayed to the next synchronization window; after the security check is passed, it sends the opening and closing commands to each electric operating mechanism in sequence according to the preset action sequence; after the action is completed, it receives the auxiliary contact status feedback and sends the execution confirmation signal back to the cloud.

[0023] Preferably, the cloud-based cluster management platform provides differentiated operation interfaces for different roles, including a cluster overview area, a real-time alarm list area, and a fault location topology area for maintenance personnel. When a fault occurs, the faulty section is automatically highlighted and a recommended power restoration path and estimated restoration time are overlaid. For equipment management engineers, there is a single-station health profile area and an aging trend curve area, which displays the historical health index curves, equivalent aging growth trends, and remaining life prediction values ​​of each transformer. For dispatch management personnel, there is a dispatch strategy recommendation area and an equipment leasing status area, which displays the equipment distribution at each stage of the cluster's life cycle, recommended load allocation schemes, their expected aging costs, and lease expiration reminder information. The equipment leasing status area and lease expiration reminder information are generated based on the health index and remaining life estimation values ​​of each transformer in the cluster health profile.

[0024] The technical effects and advantages of this invention are as follows:

[0025] 1. This invention sets up a drift detection and identification unit in the edge computing gateway, decomposes the total feature offset along three physical dimensions: temperature, aging, and load, distinguishes between normal evolution of equipment physical state and actual faults. For cases determined to be normal drift, when the drift state tends to stabilize and the local diagnostic model begins to produce a continuous low score anomaly for the normal sample under the new normal, the model update request unit uploads the drift feature vector to the cloud. The cloud generates updated individual model parameters through cluster-level correlation analysis and incremental learning and sends them back to the edge computing gateway. This allows the diagnostic model to evolve with the physical state of the equipment, rather than running continuously with a fixed model, thereby eliminating the diagnostic benchmark offset caused by seasonal changes and equipment aging, avoiding false alarms and missed alarms, and ensuring the long-term reliability and trustworthiness of the intelligent operation and maintenance system.

[0026] 2. This invention sets up a cluster health and load balancing scheduling engine in a cloud-based cluster management platform. During fault recovery and load allocation, it uses the sum of the equivalent life loss increment, which includes the traditional power scheduling target and is weighted by the differentiated aging cost weights of each transformer, as the optimization objective function. Based on the current life stage label of each transformer (healthy, mature, aging, or accelerated aging), it dynamically assigns aging cost weights. Transformers in the healthy or mature stage are given lower weights to bear more load, while transformers in the aging or accelerated aging stage are given higher weights to be subject to stricter load increment constraints. This allows mature equipment to actively bear more load, while aging equipment receives protective load limits. This achieves a shift in scheduling paradigm from power sharing to balanced life loss, slows down the overall aging process of the cluster, reduces the risk of multiple devices entering a high-fault period simultaneously, and improves the long-term power supply reliability of the cluster. Attached Figure Description

[0027] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0028] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0029] Figure 2 This is a schematic diagram of the adaptive evolution process of the diagnostic model of the present invention;

[0030] Figure 3 This is a schematic diagram of the cluster health balance scheduling process of the present invention;

[0031] Figure 4 This is a schematic diagram comparing the diagnostic accuracy of the present invention;

[0032] Figure 5 This is a schematic diagram showing the comparison of false alarm rates in this invention. Detailed Implementation

[0033] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0034] The following implementation describes an IoT-based intelligent prefabricated substation cluster management operating system. This system is applicable to the cluster management of prefabricated substations, power transformers, distribution switch control equipment, and electrical complete sets of equipment in scenarios involving distributed photovoltaic power generation, energy storage devices, charging piles, and new energy vehicle battery swapping facilities. It is also applicable to similar prefabricated substation cluster management scenarios with multi-source data sensing, edge computing processing, fault diagnosis and early warning, power supply self-healing scheduling, equipment leasing management, and full lifecycle operation and maintenance functions. The system interface may include a cluster overview area, a single station health profile area, a real-time alarm list area, a fault location topology area, a power supply recovery path display area, an aging trend curve area, an equipment leasing status area, and a scheduling strategy recommendation area. The interface directly displays mainly operational information visible to operators, such as the operating status of each transformer, real-time load rate, health index score, estimated remaining lifespan, current alarm level, equipment lease period and expiration reminder, recommended dispatching scheme, and power restoration path. The system backend further utilizes these visible data and data hidden by the interface, such as high-frequency vibration waveforms, dissolved gas component concentrations in oil, winding hot spot temperature curves, insulation resistance attenuation rate, partial discharge phase spectrum, mechanical characteristic parameters of switchgear, historical drift feature vectors, and cumulative equivalent aging amount, to complete the continuous adaptive evolution of diagnostic models, cluster lifespan loss balancing optimization, equipment lease asset health assessment, safety timing verification of fault isolation and power restoration schemes, and dynamic generation of full lifecycle operation and maintenance strategies.

[0035] In this embodiment, vibration spectrum characteristics, temperature distribution, dissolved gas composition in oil, and insulation resistance attenuation rate are the core parameters for the full life cycle health assessment of prefabricated substations. The baseline values ​​of these parameters are not fixed but continuously shift with seasonal changes, equipment aging, and fluctuations in operating conditions. For prefabricated substation clusters connected to distributed photovoltaic, energy storage, and charging piles, the operating states of individual stations vary greatly, and the aging processes are asynchronous. If the diagnostic model maintains a fixed baseline from the initial commissioning phase without adaptive adjustments, it may not only lead to a single false alarm but also cause the entire cluster's health assessment system to systematically fail within a few years. Similarly, if the scheduling strategy only... While focusing on power balance and ignoring the different aging levels of each transformer substation may seem to balance the electrical load in the short term, it accelerates the depletion of the already limited remaining lifespan of aging equipment in the long term, leading to a concentrated outbreak of clustered insulation faults. Therefore, this implementation does not treat the cloud-based collaborative operating system merely as ordinary monitoring software for displaying equipment status and issuing control commands. Instead, it is designed as a full lifecycle cluster management system that allows diagnostic models to evolve with the physical state of the equipment, proactively identifies and protects aging equipment during fault recovery, incorporates cluster lifespan balancing into scheduling optimization goals, and performs equivalent aging cost accounting for each scheduling decision.

[0036] Before describing the specific embodiments of the present invention in detail, this section first explains several basic technical concepts and conventional methods involved in the present invention to facilitate understanding by those skilled in the art. The basic technical concepts and conventional methods belong to the mature prior art in this field. The present invention proposes innovative improvements based on conventional methods.

[0037] A prefabricated substation cluster refers to a power supply network formed by interconnecting multiple prefabricated substations through distribution lines. Each prefabricated substation integrates transformers, high and low voltage switchgear, reactive power compensation devices, secondary control circuits, commercial power supply, photovoltaic power generation, energy storage devices, and electric vehicle charging facilities. Therefore, the load characteristics, power flow direction, and operating conditions of each station are significantly different. Existing cluster management methods usually rely on independent operation of single-station monitoring systems, with limited load switching and fault isolation between stations only through the dispatch center. The health status, aging degree, and remaining lifespan information of each station are isolated from each other, lacking unified full life cycle collaborative management.

[0038] An edge computing gateway is an embedded data processing device deployed on the field side of a prefabricated substation. It communicates with data acquisition units such as PLCs, protection devices, temperature sensors, vibration sensors, dissolved gas sensors in oil, and current and voltage transformers. The PLC is responsible for executing the actions of the switching equipment, the protection device is responsible for detecting fault current and outputting trip signals, the temperature sensor is responsible for collecting the hot spot temperature of the transformer winding and the ambient temperature, the vibration sensor is responsible for collecting the mechanical vibration signals of the iron core and windings, the dissolved gas sensors in oil are responsible for collecting the component concentrations of various characteristic gases in the insulating oil, and the current and voltage transformers are responsible for collecting the load current and bus voltage. The edge computing gateway undertakes data cleaning, feature extraction, local inference, and communication management functions, and uploads the processed structured data to the cloud platform through the Internet of Things communication link. In conventional edge computing solutions, the gateway only performs fixed preprocessing and data forwarding logic and does not have the ability to autonomously detect equipment state drift and update diagnostic models online.

[0039] Machine learning fault diagnosis models refer to anomaly detection and fault classification algorithms trained using historical operating data. Conventional methods employ algorithms such as deep autoencoders, isolated forests, support vector machines, or convolutional neural networks, using feature vectors such as vibration spectrum, temperature gradient, and gas component ratio in oil as input, and outputting anomaly scores or fault type discrimination results. In existing technologies, models are usually trained once in the cloud using historical data and then deployed to the field, without being updated during operation. However, for long-term operating equipment such as prefabricated substations, monitoring features will undergo irreversible shifts with seasonal environment and equipment aging, causing the distribution of input data to gradually deviate from the distribution of training data, i.e., concept drift, which ultimately leads to problems such as increased false alarm rate and increased false negative rate.

[0040] Insulation equivalent aging and health balance dispatch refers to an optimization method that uses the cumulative thermal aging of the internal insulation material of the transformer as a quantitative indicator of the equipment's health status and incorporates the differentiated aging status of each prefabricated substation into the cluster dispatch decision. Conventional dispatch methods, when restoring power supply or allocating load, usually take power balance, minimizing network loss, or maximizing recovery speed as objective functions, without considering the current aging status, hot spot temperature margin, and remaining life of each prefabricated substation. However, according to the insulation thermal aging theory, the rate of insulation life consumption doubles for every approximately 8°C increase in operating temperature. Therefore, the actual equivalent life loss of prefabricated substations that have been subjected to heavy loads, high temperatures, and impact loads for a long time is much higher than that of lightly loaded equipment. This is the core problem that health balance dispatch aims to solve. It prioritizes the allocation of loads to prefabricated substations with good health status and sufficient temperature margin, while limiting the burden of excessive incremental loads on prefabricated substations that are already in the accelerated aging stage. This slows down the overall aging process of the cluster and avoids multiple devices entering a high-failure period at similar time points.

[0041] Reference Figure 1As shown, the present invention provides a technical solution: an intelligent prefabricated substation cluster management operating system based on the Internet of Things. The system adopts a three-layer collaborative architecture of "end-edge-cloud", including: various sensors and actuators deployed inside each prefabricated substation, edge computing gateways set on the field side, and a cloud cluster management platform deployed in a remote center. The edge computing gateways and the cloud platform are connected through an Internet of Things communication network to realize data uploading and command issuance. The edge computing gateways interact with each other and make collaborative decisions through the cloud platform.

[0042] The sensing and execution units deployed within each prefabricated substation node comprise four parts: electrical quantity acquisition unit, transformer status sensing unit, environmental and auxiliary status sensing unit, and execution mechanism unit. These constitute the end-side sensing and execution layer in the three-layer collaborative architecture of this system.

[0043] The electrical quantity acquisition unit is installed in the incoming line cabinet, outgoing line cabinet, and tie cabinet. It includes current transformers, voltage transformers, and multi-functional power meters. It collects three-phase current, three-phase voltage, active power, reactive power, power factor, and frequency of each circuit in real time. Conventional box-type substations are only equipped with protection-level transformers for fault tripping. In this system, the electrical quantity acquisition unit adopts a dual configuration of high-precision metering transformers and protection-level transformers. On the one hand, it meets the wide dynamic range required for fault protection, and on the other hand, it provides high-precision power flow distribution data to the cloud dispatch engine. At the same time, the unit has a high-resolution transient waveform recording function. When a sudden current change is detected, waveform recording is automatically triggered to record holographic waveform data of several cycles before and after the fault. After these waveform data are uploaded to the edge computing gateway, they become the raw material for fault type identification based on transient characteristics. This allows fault diagnosis to no longer rely solely on the over-limit judgment of steady-state electrical quantities, but to capture the transient energy propagation characteristics and phase relationship at the moment of the fault, providing data support for distinguishing between photovoltaic weak feeder fault current and charging pile normal start-up inrush current.

[0044] The transformer condition sensing unit includes a fiber optic temperature sensor, a triaxial vibration acceleration sensor, an online dissolved gas monitoring device in the oil, and a partial discharge ultra-high frequency sensor. The fiber optic temperature sensor is embedded in the insulation pad at the end of the transformer winding and is led out to the outside of the transformer tank via an optical fiber lead to directly measure the winding hot spot temperature. It operates in a continuous online measurement mode. The temperature data is converted into a digital signal by the fiber optic demodulator and then periodically sent to the edge computing gateway. This method of directly measuring the winding temperature overcomes the inherent defects of conventional top-level oil temperature conversion methods, such as lag and large deviation during sudden load changes. When the edge computing gateway subsequently calculates the equivalent insulation aging amount, it can be based on the actual hot spot temperature rather than an indirect estimate, thus significantly enhancing the accuracy of aging assessment.

[0045] The triaxial vibration accelerometer is installed at the reinforcing rib node on the outer wall of the transformer tank. It collects vibration acceleration signals in the axial direction of the iron core, the radial direction of the winding, and the tangential direction. Its sampling range covers the complete spectrum of iron core magnetostrictive vibration, winding electromagnetic force vibration, and transient vibration of tap changer switching. The sensor generates vibration spectrum feature vectors at fixed time intervals and uploads them to the edge computing gateway. After receiving the vibration spectrum, the edge computing gateway compares it with the historical reference spectrum of the station. When a systematic shift in the peak position or amplitude of the spectrum is found, the subsequent drift identification process is initiated to determine whether the shift is related to changes in ambient temperature or equipment aging, thereby avoiding direct judgment as a fault.

[0046] The online dissolved gas monitoring device separates characteristic gases from the transformer oil tank through a breathable membrane. Employing a combined gas chromatography and semiconductor detection principle, it analyzes the concentrations of hydrogen, carbon monoxide, methane, acetylene, ethylene, and ethane in real time. The analysis results are periodically transmitted to an edge computing gateway via a fieldbus. Upon receiving the gas component data, the edge computing gateway performs multi-dimensional cross-validation with data such as vibration spectrum and temperature. When one dimension shows an abnormal trend, synchronous data from other dimensions can corroborate or rule out other possibilities. For example, if a vibration sensor detects a spectral anomaly, the edge computing gateway simultaneously checks whether there is an increase in characteristic gases in the dissolved gas mixture or whether a partial discharge sensor outputs a corresponding pulse. If there is only vibration anomaly without evidence from other dimensions, the system will not simply classify it as a fault but will mark the deviation as a drift signal to be identified—this mechanism is the core difference between this system and the existing single-dimensional over-limit alarm mode.

[0047] The UHF partial discharge sensor is installed inside the transformer tank at the window flange to capture the electromagnetic wave signal of partial discharge caused by insulation defects. It adopts a trigger-based working mode. When the amplitude of the detected partial discharge pulse exceeds the set threshold, it records the arrival time, amplitude, and phase information of the pulse, forms a partial discharge phase spectrum, and uploads it. The above four types of sensors construct a multi-physics three-dimensional perception system for transformer operating status from four physical dimensions: temperature, mechanical vibration, insulating oil chemical characteristics, and partial discharge. After the data from each dimension are aggregated at the edge computing gateway, time alignment and correlation analysis are performed. The system's judgment of equipment status is based on the cross-validation of multiple physical quantities, thus fundamentally improving the reliability and interpretability of anomaly detection.

[0048] The environmental and auxiliary status sensing unit includes an internal environmental temperature and humidity sensor, a cooling fan airflow sensor, a cable joint temperature sensor, and a switchgear contact temperature sensor. The environmental temperature and humidity sensor is installed on the partition between the high and low voltage compartments inside the transformer substation to monitor the risk of condensation inside the substation. The temperature and humidity data is uploaded to the edge computing gateway and analyzed in conjunction with the transformer load rate and cooling fan operating status to provide environmental constraints for the cooling system control strategy. The cable joint temperature sensor uses a strap-type wireless temperature measurement structure, directly attached to the terminal of the incoming and outgoing cables, and periodically sends temperature data to the edge computing gateway via a short-range wireless protocol. The switchgear contact temperature sensor... The sensor employs surface acoustic wave (SAW) passive temperature measurement technology and is installed on the contact arms of circuit breakers and the contact fingers of disconnect switches. It can monitor contact temperature online without battery power. Its operating principle utilizes the characteristic that the resonant frequency of SAW devices changes with temperature. Temperature information is read non-contactly through an external excitation signal. This solves the engineering problems of conventional active temperature sensors, such as the difficulty of battery replacement and susceptibility to electromagnetic interference under high-voltage environments. After receiving the temperature data, the edge computing gateway correlates it with the load current. When the temperature of the joint or contact rises abnormally without a significant increase in current, it can identify the trend of connection deterioration caused by increased contact resistance in advance and trigger preventive maintenance prompts.

[0049] The actuator unit includes the electric operating mechanisms for each incoming and outgoing circuit breaker, the electric operating mechanism for the tie switch, the frequency converter controller for the transformer cooling fan, and the switching controller for the reactive power compensation device. The electric operating mechanism has local and remote switching capabilities. During normal operation, it is in remote mode to receive opening and closing commands from the edge computing gateway. During local maintenance, it switches to local mode and operates via the cabinet door button, thus ensuring the safety of maintenance personnel. When the edge computing gateway sends control commands to the actuator, the actuator does not execute them immediately but first completes local safety verification: after receiving the opening and closing command, the electric operating mechanism first checks the energy storage status of its own energy storage spring. If the energy storage is insufficient, it first starts the energy storage motor, and then executes the opening and closing operation after the energy storage is completed. Before the tie switch closes, its synchronous phasor measurement module collects and detects the voltage amplitude, frequency, and phase difference of the busbars on both sides of the tie switch in real time. The closing circuit is only turned on when the three indicators of voltage difference, frequency difference, and phase difference all meet the preset synchronization conditions. This safety mechanism, which upgrades the closing conditions from static rule judgment to dynamic measurement and verification, is particularly critical in scenarios where the power flow direction changes frequently due to photovoltaic and energy storage access. The impact damage to transformers and switching equipment caused by asynchronous closing is thus effectively avoided. After the action is completed, the electric operating mechanism sends the auxiliary contact status signal back to the edge computing gateway as a confirmation feedback of the action completion, forming a closed-loop control link from command issuance to execution confirmation.

[0050] Above the edge sensing and execution layer, edge computing gateways are deployed in various prefabricated substations, acting as intermediate processing nodes between edge sensor data and the cloud platform. Their hardware platform adopts an embedded processor architecture, with built-in storage units and IoT communication modules. They interact bidirectionally with edge sensors and actuators via fieldbus or Ethernet interfaces, and connect to the cloud cluster management platform via mobile communication networks or wired private networks. In conventional edge computing solutions, the gateway only handles data transmission and fixed preprocessing logic. In this system, the edge computing gateway performs three core functions: first, it performs time alignment, feature extraction, and data quality assessment on multi-source heterogeneous sensor data from the edge, transforming raw sampled data into structured feature vectors; second, it runs a local lightweight diagnostic inference engine and drift detection module to complete real-time equipment status judgment and online concept drift identification; and third, it acts as a local safety execution agent for cloud scheduling commands, completing safety condition verification and action timing orchestration before control commands are issued to the actuators.

[0051] Multi-source data aggregation and feature extraction constitute the basic components of edge computing gateway operation. Each sensor on the edge reports data according to its own independent sampling period and communication protocol. The electrical quantity acquisition unit synchronously samples and outputs effective value parameters and transient waveform segments at the power frequency. The vibration acceleration sensor outputs spectral feature vectors at fixed time intervals. The dissolved gas monitoring device in oil outputs the concentration values ​​of six characteristic gases at fixed periods. Each temperature sensor outputs temperature measurement data at periods ranging from seconds to minutes. After receiving these heterogeneous data streams, the edge computing gateway firstly timestamps each data point and aligns it according to a unified time reference, so that electrical quantity, vibration spectrum, gas composition and temperature data at the same moment form a correlated snapshot. Subsequent multi-dimensional cross-validation and drift identification are all based on this time alignment.

[0052] In terms of feature extraction, the edge computing gateway does not simply forward the raw sampled data, but extracts features that carry physical meaning for equipment status identification and aging assessment. Taking vibration signal processing as an example, the edge computing gateway performs a fast Fourier transform on the raw waveform uploaded by the triaxial vibration accelerometer to extract the fundamental frequency amplitude, second harmonic amplitude, third harmonic amplitude, and high-frequency energy ratio, forming a vibration spectrum feature vector. This compressed representation retains the key spectrum information of iron core magnetostriction and winding electromagnetic force vibration, while significantly compressing the data transmission volume, thereby effectively alleviating the constraint of industrial field communication bandwidth bottleneck on high-frequency monitoring. For dissolved gas data in oil, the edge computing gateway calculates characteristic gas ratios based on chromatographic analysis results, covering the ratio of methane to hydrogen, acetylene to ethylene, and carbon monoxide to carbon dioxide. These ratios serve as key input features for insulation fault type identification, and the extraction process is completed locally, avoiding the transmission delay caused by uploading the entire raw chromatographic data to the cloud.

[0053] Calculating the equivalent aging amount of insulation is a crucial numerical processing step performed locally by the edge computing gateway. The thermal aging rate of insulation materials has an exponential relationship with hotspot temperature. In engineering, a simplified relationship is widely used: for every certain increase in operating temperature, the insulation lifespan consumption rate doubles. The edge computing gateway directly measures the winding hotspot temperature using fiber optic grating temperature sensors. Based on this thermal aging law, it accumulates the equivalent aging amount. One feasible calculation method expresses the aging acceleration factor as:

[0054]

[0055] In equation (1), express Aging-accelerating factors at all times This represents the measured value of the winding hot spot temperature at the current moment. Indicates the reference temperature for the lifespan of the insulation material. This represents the temperature rise required to halve the lifespan, typically 6 to 8 K for oil-immersed transformer insulation systems. After obtaining the acceleration factor, the equivalent aging amount is accumulated in hours as a statistical window:

[0056]

[0057] In equation (2), Indicates the cumulative equivalent aging amount. Indicates the first Length of each statistical window The acceleration factor corresponds to the average temperature of hotspots within the window. The edge computing gateway updates the cumulative value after each statistical window is completed and periodically uploads the converted health index to the cloud. The cloud can then directly obtain structured health parameters that can be used for scheduling and optimization without having to receive and process the high-frequency sampled raw temperature sequence. Communication overhead and cloud computing load are reduced in tandem.

[0058] The local lightweight diagnostic inference engine and drift detection module constitute the core components that distinguish the edge computing gateway from conventional solutions. In the initial stage of operation, the cloud utilizes complete annual historical monitoring data from each transformer substation to train a basic diagnostic model for the entire population. This model is constructed using an unsupervised anomaly detection algorithm, taking multi-source features such as vibration spectrum feature vectors, temperature gradient matrices, and gas component concentration ratios in the oil as inputs, and outputting anomaly scores and fault type probability distributions. The basic model is compressed and quantized and distributed to each edge computing gateway, serving as the initial state of the individual diagnostic model for that transformer substation.

[0059] During runtime, the edge computing gateway inputs time-aligned multi-source feature vectors into the local diagnostic model, performs online inference, and outputs the current anomaly score. Simultaneously, it runs an independent drift detection module to continuously monitor the statistical deviation of the model's input feature distribution from the baseline distribution at the initial deployment stage. A feasible drift detection metric is based on the Mahalanobis distance between the feature mean and the baseline mean within a sliding window.

[0060]

[0061] In equation (3), for Time drift statistics distance, This represents the mean vector of the feature vectors within the current sliding window. This represents the mean vector of baseline characteristics during the initial stage of operation. The inverse covariance matrix representing the baseline characteristic distribution is used. The preset threshold is determined based on the statistical distribution of drift distance under normal operating conditions during the initial commissioning phase. The 95th percentile of this distribution is taken as the threshold to control the misjudgment rate of concept drift under normal operating conditions. When the threshold is continuously exceeded, the module determines that the concept is drifting and triggers drift type identification, rather than directly determining that the device is faulty.

[0062] Reference Figure 2 As shown, the drift type identification process constitutes the key logic for the edge computing gateway to achieve the co-evolution of the diagnostic model with the device. After detecting concept drift, the edge computing gateway decomposes the total feature offset along multiple physical dimensions to determine the direction of the deviation. A decomposition framework divides the total offset vector... The expression is as follows:

[0063]

[0064] In equation (4), The total deviation between the current eigenvector and the baseline eigenvector. This indicates that the temperature environment affects the drift component. This indicates the component of equipment aging drift. Indicates the load change drift component. This represents the residual terms for sensor noise and unmodeled factors. The edge computing gateway monitors the temporal evolution of each component to identify the dominant factors. It is strongly correlated with the winding hot spot temperature and the ambient temperature, and exhibits seasonal periodic fluctuations. With cumulative equivalent aging It increases monotonically, in sync with the slow energy shift in certain frequency bands of the vibration spectrum. It is mapped to load rate and power factor, and experiences short-term jumps during periods of charging pile load impact or photovoltaic output fluctuations.

[0065] For the specific calculation of each drift component, a simplified linear approximation is as follows:

[0066]

[0067]

[0068]

[0069] In equations (5) to (7), and Each represents the current ambient temperature and the reference ambient temperature. Let be the temperature sensitivity coefficient vector. The cumulative equivalent aging amount is obtained from equation (2). Let be the vector of aging sensitivity coefficients. Indicates load rate, Load sensitivity coefficient vector; unit vector Dimension and eigenvectors The dimensions are consistent, and the values ​​are all A vector, indicating that the components of temperature, aging, and load contribute in the same direction to each feature dimension; Each coefficient vector was determined by fitting historical drift data accumulated after the commissioning of each transformer substation using partial least squares regression, with labels including ambient temperature, cumulative equivalent aging, and load rate. After calibration during the system commissioning phase, the data was written into the configuration parameters. or It dominates the total deviation, and If there are no significant anomalies, the edge computing gateway will classify the drift as normal drift induced by environmental factors or aging, and will not trigger a fault alarm; conversely, if the residual term... If the residual value continues to increase and cannot be explained by known environmental or aging factors, the residual is abnormally marked as a potential fault signal and enters the fault diagnosis process.

[0070] For cases determined to be normal drift, the edge computing gateway does not immediately trigger model parameter updates, but continues to observe whether the drift state tends to stabilize. When the abnormal score of the local diagnostic model output of the normal sample under the new normal continues to be lower than the preset alarm threshold, it indicates that the model benchmark has deviated substantially from the current physical state of the device. The edge computing gateway will then upload the recently accumulated normal drift feature vectors to the cloud after desensitization and encryption, and request individual model parameter updates. After receiving multiple drift reports of the same model of transformer substation in the cloud, cluster-level correlation analysis is performed to further confirm the nature of the drift. The cluster-level correlation analysis uses cosine similarity as a measure of consistency between drift feature vectors. When the cosine similarity between drift feature vectors of transformer substations of the same model, batch, and under similar conditions exceeds a preset threshold by more than a preset proportion, it is determined to be common environment drift. If only the drift vector of an individual device has a similarity to other devices in the population that is below the threshold and its correlation coefficient with its own cumulative equivalent aging amount is higher than a preset value, it is determined to be individual aging drift. For confirmed common environment drift or individual aging drift, the model parameters are updated through incremental learning. The incremental learning adopts an elastic weight consolidation method, introducing a quadratic constraint term on the network parameters related to historical fault modes into the loss function. The constraint strength is proportional to the importance of each parameter to the historical fault classification task, so that only minor adjustments are made to the key parameters for historical fault identification during the adaptation to the new normal. This allows the model to adapt to the new data distribution without forgetting the historical fault patterns. After the model parameters are updated, they are transmitted back to the edge computing gateway wirelessly to complete the evolution of the individual diagnostic model for the transformer. As a result, the system not only knows the magnitude of the deviation, but also further determines the cause and origin of the deviation. Operators no longer receive general abnormal alarms, but differentiated handling suggestions pointing to specific causes. Temperature fluctuations cause drift, which is automatically compensated by the model parameters. Aging causes drift, which updates the new normal benchmark of the transformer. Load shocks cause short-term deviations, which are decoupled from fault diagnosis. The false alarm rate is thus systematically controlled.

[0071] The secure execution agent function reflects the critical position of the edge computing gateway in the control link. After the cloud cluster management platform generates a power restoration or load transfer plan based on the cluster health and balance scheduling strategy, the control command is sent to the corresponding transformer substation edge computing gateway via the IoT communication link. After receiving the control command, the edge computing gateway does not directly forward it to the execution mechanism, but first performs local safety condition verification: On the one hand, it reads the current health index and remaining life estimate of the transformer substation. If the command requires the load increment to exceed the preset life protection threshold of the equipment, the edge computing gateway reports the rejection reason to the cloud and requests a regeneration of the scheduling plan; on the other hand, it performs transient safety pre-verification on the tie switch closing command. Using the real-time voltage amplitude, frequency and phase difference of the two buses provided by the end-side synchronization phasor measurement module, it judges the synchronous closing conditions. If the conditions are not met, the execution is delayed and waits for the next synchronization window. After the safety verification is passed, the edge computing gateway sends the opening and closing commands to the electric operating mechanism in sequence according to the preset action sequence. After the action is completed, it receives the status feedback of the auxiliary contacts and sends the execution confirmation signal back to the cloud, forming a closed-loop control from scheduling decision to secure execution.

[0072] Through the coordinated operation of the aforementioned functional modules, the edge computing gateway plays a triple role in the three-layer architecture: real-time perception and processing, local intelligent judgment, and secure execution agent. It aggregates multi-source sensor data from the edge side downwards to complete feature extraction and aging assessment; it runs a local diagnostic engine and drift detection module internally to achieve adaptive model evolution; and it receives cloud scheduling instructions upwards to complete security verification and timing control. This design allows the system to maintain basic fault detection and security control capabilities even in extreme cases of communication link interruption. The edge computing gateway continues to detect anomalies based on the local model. Once a serious fault characteristic is identified, it autonomously executes a preset emergency isolation strategy without relying on cloud confirmation. After communication is restored, it synchronizes the offline operation records and event logs to the cloud. The system's robustness and real-time response are thus doubly guaranteed.

[0073] Above the edge-side intelligent processing layer, the cloud-based cluster management platform, located on a remote data center server or cloud host, acts as the central brain of the entire system. It maintains long-term connections with the edge computing gateways of each prefabricated substation via an IoT communication network, and provides a visual operation interface and data query interface for operation and maintenance personnel. This constitutes the cloud-side decision-making management layer in the three-layer collaborative architecture of this system. While conventional substation monitoring systems only handle data display and historical storage on the cloud, this system's cloud platform undertakes four core functions: First, it aggregates structured health status data from all prefabricated substations within the cluster, establishing a full lifecycle health profile of the cluster; second, it runs a cluster-level scheduling optimization engine, introducing differentiated aging cost constraints in fault recovery and load allocation decisions to generate a healthy and balanced scheduling scheme; third, it maintains the basic diagnostic model of the population, receiving drift reports uploaded by each edge gateway, performing cluster-level correlation analysis and incremental learning, and wirelessly distributing individual model update parameters; fourth, it provides operation interfaces for different roles, visually presenting equipment health status, recommended scheduling schemes, and aging trend predictions to assist operation and maintenance decisions.

[0074] Cloud-based data aggregation and cluster health profiling form the foundation of the platform's operational data. Each transformer substation's edge computing gateway uploads structured operational data at fixed intervals. This data does not contain raw high-frequency sampled waveforms; it is a compressed representation after edge-side feature extraction and aging calculation. A single reported data packet includes: transformer substation identification, health index, cumulative equivalent aging amount, estimated remaining lifespan, current load rate, winding hotspot temperature, most recent drift detection status code, and abnormal event markers. After receiving the data, the cloud stores it in a time-series database to accumulate historical samples for subsequent trend analysis and model training. On the other hand, it maps the data of each node to a geographic information map or single-line map according to the cluster topology, generating an intuitive cluster health heatmap.

[0075] The cluster health profile is constructed with the equivalent aging amount of each transformer as the core indicator. The equivalent aging amount is accumulated by the edge computing gateway according to formula (2) and uploaded periodically. The cloud performs horizontal comparison and group management of the aging amount of all equipment in the cluster: transformers in the early stage of operation and with low cumulative equivalent aging amount are marked as healthy period; those with cumulative equivalent aging amount reaching a certain proportion of the design life are marked as mature period; those approaching the critical value of insulation life are marked as aging period; those with latent defects detected or cumulative aging amount exceeding the warning threshold are marked as accelerated aging period. This group label is a non-static attribute and is dynamically refreshed with the health index uploaded by the edge computing gateway. Transformers at different life stages are assigned different load-bearing weights and life protection thresholds in subsequent scheduling optimization, thereby providing a quantitative basis for healthy balanced scheduling.

[0076] The cluster-level maintenance and evolution of the population-based diagnostic model constitutes a key function that distinguishes this cloud platform from conventional cloud platforms. In the initial stage of system operation, the cloud uses complete annual historical monitoring data from each transformer substation to train the population-based model. This model is built based on an unsupervised anomaly detection algorithm, with inputs including multi-source features such as vibration spectrum feature vectors, temperature gradient matrices, and oil gas component concentration ratios. It outputs anomaly scores and fault type probability distributions. The basic model is compressed and quantized, then distributed to each edge computing gateway as the initial state of individual diagnostic models. During daily operation, when an edge computing gateway of a transformer substation detects continuous and stable drift and uploads the drift feature vector, the cloud does not simply update the model for that substation individually. Instead, it initiates a cluster-level correlation analysis program. This cluster-level correlation analysis uses cosine similarity as a measure of consistency between drift feature vectors. It searches for other transformer substations of the same model, batch, and operating under similar environmental conditions. When the cosine similarity between drift feature vectors of transformer substations of the same model, batch, and operating under similar conditions exceeds a preset threshold by more than a preset proportion, it is confirmed as a common environmental drift. The cloud then applies this to the basic model. Domain-adaptive training allows the model as a whole to adapt to such environmental changes. If only individual devices exhibit shifts, and their drift vectors have a similarity to other devices in the population below a threshold and a correlation coefficient with their own cumulative equivalent aging amount above a preset value, they are identified as individual aging drift. The cloud platform generates individual model parameters for this transformer substation through incremental learning based on the basic model. Incremental learning employs an elastic weight consolidation method, introducing a secondary constraint term for network parameters related to historical fault modes into the loss function. The constraint strength is proportional to the importance of each parameter to the historical fault classification task, ensuring that only minor adjustments are made to key parameters for historical fault discrimination during the adaptation to the new normal. This allows the model to adapt to the new data distribution without forgetting historical fault modes. The updated model parameters are wirelessly transmitted back to the corresponding edge computing gateway, completing the individual model evolution. The advantages of this mechanism are: common environmental changes are uniformly adapted by the population model, avoiding the risks of sample sparsity and overfitting caused by independent training of each device; individual aging differences are finely adjusted through incremental learning, completing personalized adaptation of diagnostic benchmarks for devices at different aging stages.

[0077] Reference Figure 3 As shown, the cluster health and balance scheduling engine constitutes the core decision-making module of this cloud-based cluster management platform, and is also the key difference between this system and existing power balance-oriented scheduling schemes. When a fault occurs in the upper-level power grid and load transfer is required, or when planned load adjustments are performed, the scheduling engine initiates optimization calculations. The optimization objective function includes two terms: the first is the traditional objective, namely maximizing the power supply restoration range or minimizing network loss; the second is the aging cost term introduced by this system. The physical meaning of the aging cost term is the sum of the equivalent lifetime loss increments generated after each transformer takes on additional loads. Its calculation relies on the health index and winding hotspot temperature data uploaded by the edge computing gateway. The optimization problem can be expressed as:

[0078]

[0079] In equation (8), This represents the traditional power scheduling objective function. For the first Tabletop transformer in solution The equivalent aging increment is generated below. For the first The aging cost weighting of each transformer substation is dynamically assigned by the cloud based on its current life stage label. Substations in their healthy or prime years are assigned a lower weighting and bear more load; substations in their aging or accelerated aging stages are assigned a higher weighting and are subject to stricter load increment constraints. The relative sizes between them reflect the differences in the degree of equipment aging, and their absolute values ​​are normalized before optimization. and The trade-off coefficient between the two sub-objectives is adjusted by the operation and maintenance management personnel in the system configuration interface according to the power supply reliability requirements and asset preservation strategy, ensuring that both are satisfied. The normalization constraints are adjusted by operation and maintenance personnel in the system configuration interface according to power supply reliability requirements and asset preservation strategies.

[0080] In the scheduling optimization solution process, several constraints are introduced in the cloud: First, the load rate of each transformer substation does not exceed its current health status allowable load limit, which is estimated and uploaded in real time by the edge computing gateway based on the winding hot spot temperature margin; second, the load transfer path must ensure no islanding and no reverse power flow exceeding the limit; third, the number of tie switch operations does not exceed the mechanical life limit of the switch within a preset period. Under the above constraints, the optimization solver searches for the power restoration scheme or load allocation scheme that minimizes the objective function, generating a complete scheduling plan covering the switch operation sequence, the target load value of each transformer substation, and the expected life loss. Once the solution is completed, the scheduling plan is pushed to the operation interface in the form of a recommendation. After the operator reviews and confirms it, it is issued for execution. If the system is in fully automatic operation mode, the scheduling plan is executed automatically within the operator's preset authority range without manual confirmation. This scheduling mechanism explicitly incorporates aging costs into the optimization decision, enabling the scheduling logic to complete the paradigm shift from power amortization to balanced lifespan loss: in a single fault recovery, the equipment in its prime actively takes on more load, while the equipment in the accelerated aging period is given protective load limits. The overall aging process of the cluster is thus systematically delayed, and the risk of multiple devices entering the high-incidence period of failure at similar time points is significantly reduced.

[0081] The cloud platform's user interface provides differentiated information views and interactive functions for different roles. For maintenance personnel, the interface focuses on the cluster overview area, displaying the operating status of each transformer, real-time load rate, and current alarm level. When a fault occurs, it automatically switches to the fault location topology area, highlighting the fault section and overlaying the recommended power restoration path and estimated restoration time. Operators can confirm execution with one click or manually adjust before execution. For equipment management engineers, the interface emphasizes the single-station health profile area and aging trend curve area, displaying the historical health index curve, equivalent aging growth trend, and remaining life prediction value of each transformer, assisting in the formulation of annual maintenance plans, equipment replacement budgets, and equipment lease expiration assessments. For dispatch management personnel, the interface emphasizes the dispatch strategy recommendation area and equipment lease status area, displaying the distribution of equipment at each life stage within the current cluster, monthly life loss statistics, recommended load allocation schemes and their expected aging costs, lease expiration reminders, and comprehensive equipment health rankings, etc. The system backend further utilizes hidden data not visible on the operation interface—including historical drift feature vector sequences of each transformer, equivalent aging cost accounting records of each scheduling decision, partial discharge phase spectrum and fault association mode, cumulative number of actions and mechanical characteristic decay trend of each switchgear—to complete continuous model evolution, iterative optimization of scheduling strategies, mining of hidden fault modes, and dynamic generation of full life cycle operation and maintenance strategies.

[0082] Through the coordinated operation of the aforementioned functional modules, the cloud-based cluster management platform plays a core role in the three-layer architecture of "global data aggregation, intelligent scheduling decision-making, and model cluster evolution": Upward, it provides multi-level visual views for operation and maintenance management, from single-site health assessment to cluster aging trend prediction; downward, based on real-time health status data uploaded by each edge gateway, it runs a scheduling optimization engine that considers differentiated aging costs to generate a healthy and balanced load allocation scheme; inward, it maintains the basic diagnostic model of the population, receives drift reports, performs cluster-level correlation analysis and incremental learning, and completes the continuous evolution of individual model parameters. Between the cloud and the edge, a closed loop of model evolution is formed through edge drift report upload—cloud correlation analysis and incremental learning—model parameter download; a closed loop of scheduling control is formed through edge health data upload—cloud scheduling optimization solution—control command issuance—edge safety verification and execution feedback. The coordinated operation of these two closed loops upgrades the entire system from traditional substation monitoring with post-event alarms and manual scheduling to full lifecycle proactive management and cluster collaborative optimization.

[0083] The overall operation process of the system is as follows: After the system is powered on, the sensors on the end side of each box-type substation enter the autonomous acquisition state. The edge computing gateway synchronously receives multi-source data such as electrical quantity, vibration spectrum, dissolved gas components in oil, and temperature, completes time alignment and feature extraction, and starts the local diagnostic model to implement online anomaly detection. At the same time, the edge computing gateway calculates the aging acceleration factor corresponding to the winding hot spot temperature according to a fixed period, accumulates the equivalent aging amount according to formula (1) and formula (2), and periodically uploads the structured health data to the cloud platform.

[0084] After receiving the health index, cumulative equivalent aging amount and remaining life estimate uploaded by each transformer, the cloud platform constructs a health heat map according to the cluster topology, and marks each transformer with a life stage label - healthy period, mature period, aging period or accelerated aging period - and updates it dynamically with data refresh. Under normal operating conditions, the edge computing gateway continuously monitors the distribution of input features of the diagnostic model. If the drift detection module finds that the statistical distance exceeds the preset threshold according to formula (3), it triggers drift type identification: according to formula (4), the total offset is decomposed into temperature drift component, aging drift component and load drift component. Using equations (5) to (7) to identify the dominant factors, if it is determined to be a normal drift induced by the environment or aging and the state tends to be stable, the edge computing gateway will upload the drift feature vector to the cloud after desensitization and encryption. The cloud will start cluster-level correlation analysis - search for the drift patterns of other transformer substations of the same model, batch, and similar environmental conditions. If the common environmental drift is confirmed, the population basic model will be performed on the domain adaptive training. If the individual aging drift is confirmed, the individual model parameters of the transformer substation will be generated through incremental learning. After the model is updated, it will be wirelessly transmitted back to the edge computing gateway to complete the continuous evolution of the diagnostic model.

[0085] When a transformer fails or the upstream power grid experiences a power outage, the cloud-based scheduling engine initiates optimization calculations: it reads the current health index and winding hot spot temperature margin of each transformer, constructs a dual-objective optimization function containing traditional power targets and aging cost items according to formula (8), assigns higher aging cost weights to transformers in the aging period or accelerated aging period, applies load rate limits, no islanding formation and switching action number constraints, and solves to generate a healthy and balanced power restoration or load transfer scheme. The scheme is pushed to the operation interface in the form of a recommendation, and maintenance personnel can confirm it with one click or adjust it manually. In fully automatic mode, the scheme is automatically distributed within the preset permission range.

[0086] After receiving the cloud-based scheduling command, the edge computing gateway first performs local security checks: on the one hand, it verifies whether the load increment required by the command exceeds the preset life protection threshold of the transformer. If it does, it sends a rejection message to the cloud and requests a regeneration of the scheme; on the other hand, it performs synchronization condition checks on the closing command of the tie switch, using the synchronization phasor measurement module to provide real-time voltage amplitude, frequency, and phase difference of the two busbars. If the conditions are not met, the execution is delayed until the next synchronization window. After the security check is passed, the opening and closing commands are sent to the electric operating mechanism in sequence according to the preset timing. After the action is completed, the gateway receives the status feedback of the auxiliary contacts and sends the execution confirmation back to the cloud, forming a closed loop from scheduling decision to safe execution.

[0087] After the scheduling is completed, the edge computing gateways of each transformer substation continuously monitor the load response and temperature changes, update the health index and remaining life estimate and upload them to the cloud. The cloud updates the cluster health profile, completing a complete closed loop of perception-diagnosis-scheduling-execution-feedback. The system then returns to normal monitoring status and waits for the next round of event triggering.

[0088] Example 1

[0089] This embodiment corresponds to the daily operation monitoring and alarm handling scenario of a prefabricated substation cluster. The maintenance personnel confirm on the cloud platform interface that there are no unprocessed alarms in the cluster overview area, and that the operating status, real-time load rate, and health index score of each prefabricated substation are all within the normal range. The interface shows that the current load rate of a prefabricated substation in a residential community is low and the health index score is high, indicating that it is in its prime. Another prefabricated substation in a commercial center has a slightly lower health index score but is still in the healthy period due to the periodic increase in winding hot spot temperature caused by the load fluctuation of the charging piles.

[0090] During the continuous operation of the system background, each transformer substation's edge computing gateway receives electrical quantities, vibration spectrum, dissolved gas components in oil, and temperature data at fixed intervals. After time alignment, the data is input into the local diagnostic model for online anomaly detection. Simultaneously, based on the cumulative equivalent aging amount calculated using equations (1) and (2), the health index and remaining lifespan estimate are periodically uploaded to the cloud. The cloud refreshes the health heatmap according to the cluster topology, dynamically updating the life stage label for each transformer substation. When the hot spot temperature of a transformer substation's winding approaches the warning value, a yellow warning item automatically pops up in the real-time alarm list area of ​​the interface. Maintenance personnel click on the warning item. Following the entry, the single-station health profile area displays the recent hot spot temperature curve, vibration spectrum shift trend, and changes in gas composition in the oil of the transformer. Based on the results of cross-validation of multiple physical quantities, the system gives the judgment that "the current temperature rise is mainly caused by the increase in load, and no abnormal insulation characteristics have been detected." Based on this, maintenance personnel decide to strengthen observation or arrange load reduction during off-peak hours. In this embodiment, the daily operation of maintenance personnel is still to check the interface and handle alarms, but the system background has completed multi-dimensional cross-validation and drift identification, avoiding frequent false alarms caused by single-dimensional over-limit alarms, and improving the reliability and handling efficiency of alarms.

[0091] Example 2

[0092] This embodiment corresponds to the adaptive adjustment scenario of the diagnostic model under low temperature in winter or high temperature in summer. During the winter cold wave, the ambient temperature of a certain open-air transformer substation suddenly dropped to below zero, the viscosity of the transformer insulating oil increased significantly, and the peak position and amplitude of the vibration spectrum collected by the triaxial vibration acceleration sensor were systematically offset compared with the summer training benchmark. The edge computing gateway drift detection module calculated according to formula (3) and found that the statistical distance continued to exceed the threshold, triggering the drift type identification process.

[0093] The edge computing gateway decomposes the total offset vector into temperature drift component, aging drift component and load drift component according to equation (4). It calculates the temperature drift component using equation (5) and finds that it is strongly correlated with the winding hot spot temperature and the ambient temperature and is dominant. There is no abnormal growth of dissolved gas in the oil and the partial discharge sensor does not output the corresponding pulse. The edge computing gateway judges it as normal drift induced by the environment and does not trigger a fault alarm. After the drift state is observed to stabilize, the recently accumulated drift feature vector is desensitized and encrypted and uploaded to the cloud. The cloud searches for other transformers of the same model and batch that are operating under similar climatic conditions. It finds that multiple devices show the same spectrum offset mode and confirms that it is a common environmental drift. It performs domain adaptive training on the population basic model so that the model as a whole adapts to the low temperature operation characteristics in winter. After updating, the model parameters are wirelessly transmitted back to the edge computing gateway to complete the seasonal adaptive adjustment of the individual diagnostic model of the transformer. In this embodiment, the vibration spectrum offset caused by the low temperature in winter is not misjudged as a winding loosening fault. The system automatically completes the seasonal migration of the diagnostic benchmark. The maintenance personnel do not need to manually modify the alarm threshold. Figure 5 The diagram shows a comparison of the false alarm rates of the existing single-dimensional over-limit alarm method and the multi-dimensional cross-validation method of the present invention under different operating conditions. It can be seen that the false alarm rate of the method of the present invention remains at a low level under all operating conditions.

[0094] Example 3

[0095] This embodiment corresponds to the scenario of gradual aging tracking and diagnostic benchmark update after long-term operation of equipment. A transformer substation in an industrial park has been in operation for many years. The degree of polymerization of the insulating paper gradually decreases with the years of operation, the iron core clamps become slightly loose, the energy of certain frequency bands of the vibration spectrum increases monotonically with the cumulative equivalent aging amount, and the ratio of dissolved gases carbon monoxide to carbon dioxide in the oil slowly increases. The edge computing gateway calculates the aging drift component according to formula (6) and finds that it increases synchronously with the cumulative equivalent aging amount and the trend of change is stable. The load drift component has no significant abnormality and is judged to be normal drift induced by aging.

[0096] The edge computing gateway does not immediately trigger model parameter updates. It continuously observes the drift state until it stabilizes and the local diagnostic model begins to produce consistently low scores for normal samples under the new normal. Then, it anonymizes and encrypts the drift feature vectors before uploading them to the cloud. Cloud searches reveal that only this transformer substation exhibits this type of shift, which is strongly correlated with the cumulative equivalent aging amount, confirming it as individual aging drift. Based on the basic model, incremental learning is used to generate individual model parameters for this transformer substation. Without forgetting historical fault modes, the model adapts to the new normal state under the current aging condition. After the updated model parameters are sent back to the edge computing gateway, the abnormal score returns to the normal range. In this embodiment, the feature shift caused by the slow aging of the equipment is not continuously marked as abnormal. The system automatically completes a lifetime progressive update of the diagnostic benchmark, avoiding lifetime model failure due to natural equipment aging. Figure 4 The diagram shows a comparison of the diagnostic accuracy of the existing fixed diagnostic model and the adaptive evolutionary model of the present invention under different operating years. It can be seen that the accuracy of the model of the present invention does not decrease monotonically with the operating years.

[0097] Example 4

[0098] This embodiment corresponds to the scenario of restoring healthy and balanced power supply after a fault occurs in a transformer substation within the cluster. A transformer substation in a commercial center has an overheating fault caused by the continuous increase in the contact resistance of the cable joint. The upper protection trips, and all the charging piles and commercial loads connected to the transformer substation lose power. After receiving the fault signal, the cloud scheduling engine starts the optimization calculation: reads the current health index, winding hot spot temperature margin and remaining life estimate of the other transformer substations in the cluster, and constructs a dual-objective optimization function according to formula (8). The traditional power objective is to maximize the power supply restoration range. The weight of each transformer substation in the aging cost item is dynamically assigned according to the life stage label - the transformer substation in the adjacent residential area in the prime of life is given a lower aging cost weight, and the other old transformer substation in the aging period is given a higher weight.

[0099] The optimization solver searches for the load transfer scheme that minimizes the objective function under constraints of load rate limits, no islanding, and limited switching operations. This generates a power restoration scheme that prioritizes transferring load from faulty transformers to active transformers while limiting incremental load on aging transformers. After the scheme is pushed to the operation interface, maintenance personnel confirm and execute it. The edge computing gateway receives the command, completes security verification and synchronization detection, and controls the tie switch to close, restoring power to the non-faulty sections. In this embodiment, active transformers proactively take on more load, while aging transformers receive protective load limits, thus slowing down the overall aging process of the transformer cluster.

[0100] Example 5

[0101] This embodiment corresponds to asset health assessment and expiration management in the equipment leasing business scenario. The company leases multiple prefabricated substations to a construction project. The lease term is billed annually. The health status of the equipment directly affects the residual value assessment and renewal decision after the lease term expires. The cloud platform's equipment leasing status area displays the health index score, cumulative equivalent aging amount, estimated remaining life value, and lease expiration reminder for each leased prefabricated substation.

[0102] A leased transformer substation was connected to temporary construction loads for an extended period during the lease term, resulting in significantly higher load rates and hotspot temperatures than in normal operating scenarios. The cumulative equivalent aging of the edge computing gateway increased much faster than other transformer substations in the same batch. Before the lease expired, the system automatically generated a health assessment report for the substation, including the equivalent aging amount for each month during the lease term, the percentage of insulation life consumed, the accelerated aging factor relative to the design life, and recommended maintenance or replacement plans. Based on this report, the company negotiated with the lessee on adjustments to the lease renewal fee or equipment replacement plans. In this embodiment, the equipment leasing business was upgraded from extensive time-based billing to refined asset management based on health status, quantifying and tracing the impact of abnormal operating conditions on equipment lifespan during the lease term.

[0103] Example 6

[0104] This embodiment corresponds to the scenario of annual maintenance plan preparation and resource optimization and allocation for the cluster. The equipment management engineer can view the cluster aging trend curve area and equipment distribution statistics at each stage of life on the cloud platform. Based on the decay rate of the health index of each transformer and the predicted value of the remaining life, the system automatically generates the maintenance priority ranking and spare parts demand prediction for the next year.

[0105] Engineers discovered that the estimated remaining lifespan of three aging transformer substations was approaching critical values ​​around the same time point. Based on the current attenuation rate, they were predicted to enter accelerated aging in the near future, which would put pressure on both the shortage of maintenance resources and the reliability of power supply. Using a scheduling strategy recommendation area simulation tool, engineers pre-transferred part of the load on the circuit containing one of the transformer substations to equipment in its prime, reducing the subsequent operating load rate of that substation and slowing down its remaining lifespan consumption rate. This increased the time interval between the three devices entering accelerated aging. In this embodiment, maintenance planning was upgraded from experience-driven to data-driven, and the asynchronous scheduling of the aging process of equipment within the cluster enabled the rational allocation of maintenance resources, avoiding passive emergency repairs caused by the concentrated failure of multiple devices.

[0106] Through the modules and embodiments described above, the system, without departing from the basics of the actual operation and maintenance interface, further expands equipment status monitoring, alarm handling, scheduling operations, and asset management into multi-dimensional cross-validation, drift identification and adaptive evolution of diagnostic models, health-balanced scheduling, and full lifecycle operation and maintenance decision-making. Overall, the actual display content of the system is still consistent with the common transformer substation cluster monitoring interface, including cluster overview, single-station health status, alarm list, and scheduling scheme. However, its backend control logic has been upgraded from simple monitoring to a full lifecycle closed-loop control scheme that includes lifelong learning of diagnostic models, cluster lifecycle balanced scheduling, fault-safe self-healing, and asset health management.

[0107] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. An intelligent prefabricated substation cluster management operating system based on the Internet of Things, characterized in that, include: The sensing and execution module deployed inside each prefabricated substation includes sensors for collecting equipment operating status parameters and actuators for performing opening and closing operations. The edge computing gateway, installed on the field side of each prefabricated substation, includes a local diagnostic inference engine, a drift detection and identification unit, and a model update request unit. The local diagnostic inference engine loads an individual diagnostic model, performs online inference on the feature vectors extracted from multi-source data collected by sensors, and outputs anomaly scores. The drift detection and identification unit continuously monitors the statistical deviation of the distribution of the feature vectors relative to the baseline distribution at the initial stage of operation. When conceptual drift is detected, the total feature offset is decomposed into temperature-affected components, aging-affected components, and load change components to identify the dominant drift factors. For drifts determined to be caused by the normal evolution of the equipment's physical state, when the drift state tends to stabilize and the anomaly score output by the local diagnostic inference engine for the normal sample under the new normal continues to be lower than the preset alarm threshold, the model update request unit uploads the drift feature vectors to the cloud. The cloud-based cluster management platform deployed in a remote center includes a model evolution unit and a cluster health and balance scheduling engine; the model evolution unit receives the drift feature vector, generates updated individual model parameters through cluster-level correlation analysis and incremental learning, and sends them to the corresponding edge computing gateway; The cluster health and load balancing scheduling engine includes a bi-objective optimization solution submodule, which generates a scheduling scheme by optimizing the objective function during fault recovery and load distribution. The optimization objective function is: in, For traditional power scheduling objectives, For the first Tabletop transformer in solution The equivalent aging increment generated below For the first The aging cost weight of each transformer substation is dynamically assigned by the cloud based on the current life stage label of each substation. Substations in the healthy or prime stage are given a lower weight to bear more load, while substations in the aging or accelerated aging stage are given a higher weight to be subject to stricter load increment constraints. and The trade-off coefficient between the two sub-objectives satisfies The normalization constraint.

2. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 1, characterized in that: The sensors include sensors for measuring winding hot spot temperature, sensors for acquiring vibration spectrum, sensors for online monitoring of dissolved gas components in oil, and sensors for capturing partial discharge signals; the actuators include electric operating mechanisms for incoming and outgoing line circuit breakers and electric operating mechanisms for tie switches; the edge computing gateway also includes a data aggregation and feature extraction unit, which receives multi-source data output by the sensors, performs time alignment and feature extraction, and converts the raw sampled data into structured feature vectors before inputting them into the local diagnostic inference engine.

3. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 2, characterized in that: The drift detection and identification unit calculates the drift statistical distance based on the Mahalanobis distance between the mean of features within the sliding window and the mean of the baseline features in the initial stage of operation. The preset threshold is determined based on the 95th percentile of the statistical distribution of the drift statistical distance under normal operating conditions in the initial stage of operation. When the drift statistical distance continuously exceeds the preset threshold, it is determined that conceptual drift has occurred and the drift type identification process is triggered, rather than directly determining it as a device failure.

4. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 3, characterized in that: The drift detection and identification unit decomposes the total feature offset into drift components affected by temperature environment, drift components affected by equipment aging, drift components affected by load change, and residual terms of sensor noise and unmodeled factors. By monitoring the time evolution of each component, the dominant drift factor is identified to determine whether the drift belongs to the normal evolution of the equipment's physical state. The temperature-related drift component is related to the winding hot spot temperature and the ambient temperature and exhibits seasonal fluctuations. The equipment aging drift component increases monotonically with the cumulative equivalent aging amount. Both correspond to the normal evolution of the equipment physical state described in claim 1. The load change drift component is mapped to the load rate and power factor and exhibits short-term jumps during periods of impact load. When the temperature-related drift component or the equipment aging drift component dominates the total deviation and the load change drift component shows no significant abnormalities, it is determined to be normal drift and does not trigger a fault alarm. When the residual term after decomposition continues to increase and cannot be explained by known environmental or aging factors, it is marked as a potential fault signal and enters the fault diagnosis process.

5. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 1, characterized in that: The cloud-based cluster management platform also includes a cluster health profiling unit, which collects structured health data uploaded by edge computing gateways of each transformer substation and generates a cluster health profile containing life stage tags for each transformer substation. After receiving the drift feature vector, the model evolution unit retrieves the drift feature vectors of other transformer substations of the same model, batch, and operating under similar environmental conditions, and compares them using cosine similarity as a consistency measure. If the cosine similarity between the drift feature vectors of multiple devices under similar environmental conditions is higher than a preset threshold, it is confirmed that they are common environmental drifts, and domain adaptive training is performed on the population basic model. If only a few devices show a shift and it is strongly correlated with the cumulative equivalent aging amount, it is confirmed as individual aging drift. Incremental learning is carried out on the basis of the basic model through the elastic weight consolidation method. A secondary constraint term for the network parameters related to the historical failure mode is introduced into the loss function to generate the individual model parameters of the transformer. The model adapts to the new data distribution without forgetting the historical failure modes.

6. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 5, characterized in that: The life stage labels are dynamically refreshed by the cloud based on the health index and cumulative equivalent aging amount periodically uploaded by the edge computing gateway of each transformer. Transformers that are in the early stage of operation and have a low cumulative equivalent aging amount are marked as healthy. Transformers with a cumulative equivalent aging amount that has reached a certain proportion of the design life are marked as mature. Transformers that are close to the critical value of insulation life are marked as aging. Transformers that have detected latent defects or whose cumulative aging amount exceeds the warning threshold are marked as accelerated aging.

7. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 1, characterized in that: The edge computing gateway also includes an aging quantization unit, which calculates the aging acceleration factor based on the insulation thermal aging law using the winding hot spot temperature, accumulates the equivalent aging amount in hourly statistical windows, and periodically uploads the converted health index to the cloud cluster management platform, so that the cloud can directly obtain structured health parameters without receiving the original temperature sequence of high-frequency sampling; the equivalent aging increment in the objective function is calculated based on the accumulated equivalent aging amount.

8. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 7, characterized in that: The cluster health balancing scheduling engine also includes a constraint application submodule. The constraints applied during the optimization process include: the load rate of each transformer does not exceed the upper limit of its current health status, which is estimated and uploaded in real time by the edge computing gateway based on the winding hot spot temperature margin; the load transfer path ensures that there is no islanding and no reverse power flow exceeding the limit; and the number of times the tie switch operates does not exceed the mechanical life limit of the switch within a preset period.

9. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 1, characterized in that: After receiving the scheduling instruction from the cloud cluster management platform, the edge computing gateway first performs local security condition verification. The system reads the current health index and estimated remaining lifespan of the transformer. If the load increment required by the instruction exceeds the preset lifespan protection threshold of the equipment (the lifespan protection threshold is the maximum load increment corresponding to the upper limit of the allowed winding hot spot temperature under the current health state of the transformer), it sends a rejection reason to the cloud and requests a regeneration of the scheduling scheme. For the tie switch closing instruction, it uses the real-time voltage amplitude, frequency, and phase difference of the two busbars provided by the synchronization phasor measurement module set at the tie switch to perform synchronization condition detection. If the condition is not met, it is delayed until the next synchronization window. After the safety verification is passed, it sends the opening and closing instructions to each electric operating mechanism in sequence according to the preset action sequence. After the action is completed, it receives the auxiliary contact status feedback and sends the execution confirmation signal back to the cloud.

10. The IoT-based intelligent prefabricated substation cluster management operating system according to claim 1, characterized in that: The cloud-based cluster management platform provides differentiated operation interfaces for different roles, including a cluster overview area, a real-time alarm list area, and a fault location topology area for operation and maintenance personnel. When a fault occurs, the faulty section is automatically highlighted and a recommended power restoration path and estimated restoration time are overlaid. For equipment management engineers, there is a single-station health profile area and an aging trend curve area, which displays the historical health index curves, equivalent aging growth trends, and remaining life prediction values ​​of each transformer. The scheduling strategy recommendation area and equipment lease status area for scheduling managers display the distribution of equipment at each stage of the cluster's life cycle, recommended load allocation schemes, their expected aging costs, and lease expiration reminder information. The equipment lease status area and lease expiration reminder information are generated based on the health index and remaining life estimate of each transformer in the cluster's health profile.

Images