A multi-physical field coupling cloud edge coordination low-voltage power distribution cabinet whole life cycle intelligent operation and maintenance system

Abstract

This invention discloses a multi-physics field coupled cloud-edge collaborative intelligent operation and maintenance system for low-voltage distribution cabinets throughout their entire lifecycle. Belonging to the field of intelligent operation and maintenance technology for power distribution equipment, the system comprises a four-layer architecture: a cabinet-side perception layer, an edge computing gateway layer, a cloud-based digital twin platform layer, and a mobile terminal layer. The cabinet-side system synchronously collects multi-dimensional physical signals from electrical, mechanical, insulation, and environmental dimensions. The edge gateway performs multi-protocol standardization conversion, multi-sensor time-series alignment and noise reduction, and adaptive weighted coupling fault feature extraction based on an 80% rated load boundary. Locally, it achieves graded early warning and fault isolation within 100ms and supports 7 days of offline caching and breakpoint resumption. The cloud assigns a unique digital identity to each low-voltage distribution cabinet. During daily low-load periods, a gradient descent algorithm is used to achieve timed self-calibration of the digital twin model. LSTM network combined with an 85% graph similarity threshold is used to trace the root cause of faults. When the similarity is less than 85%, unknown fault samples are automatically retained. Maintenance data uploads instantly update the graph and iterate the model, forming a complete operation and maintenance closed loop. This invention addresses the shortcomings of existing technologies, such as single monitoring dimensions, delayed cloud processing, distortion of static twin models, low accuracy of fault diagnosis, and fragmented operation and maintenance processes. It proposes an innovative overall solution with multi-module collaborative coupling, which solves the industry pain points of high false alarm and false alarm rates, delayed fault handling, decreased model accuracy, and inability to iterate operation and maintenance independently in traditional low-voltage distribution cabinet operation and maintenance. It can identify hidden equipment faults in advance, significantly improve power supply reliability, reduce manual operation and maintenance costs, and is applicable to the intelligent management and control of low-voltage distribution cabinets in various power distribution rooms.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and operation and maintenance technology for low-voltage complete sets of power distribution equipment. Specifically, it relates to an intelligent management and control system for low-voltage distribution cabinets that integrates multi-physical field data adaptive fusion, edge hierarchical interlocking protection, dynamic self-calibration digital twin, and full-process closed-loop operation and maintenance. It is applicable to multi-dimensional status perception, on-site fault handling, dynamic simulation of digital twin, and full-process predictive operation and maintenance management of low-voltage distribution cabinets in industrial plants, commercial complexes, and park power distribution rooms. Background Technology

[0002] Low-voltage switchgear, as the core equipment for power distribution in low-voltage power distribution systems, plays a crucial role in power transfer, distribution, and protection. During long-term operation, it is prone to hidden defects such as insulation degradation, loose contact, mechanical opening and closing faults, and partial discharge. If these hidden dangers are not identified and addressed in advance, they can easily lead to safety accidents such as short circuits, electrical fires, and unplanned power outages, causing significant economic losses to production and operations. Currently, several intelligent monitoring and maintenance solutions for low-voltage switchgear have been publicly disclosed in the industry, capable of basic data collection and simple alarm functions. However, a review of existing core comparative documents reveals that current technologies still have many inherent technical defects that cannot be avoided, failing to meet the requirements for high-precision, low-false-alarm, fully automated, and full-cycle intelligent operation and maintenance.

[0003] Existing mainstream technical solutions can be mainly divided into four categories, and their specific technical shortcomings are as follows: The first category is cloud-based centralized electrical parameter monitoring solutions. A representative example is the cloud-based remote operation and maintenance management and control system for high and low voltage distribution cabinets disclosed in CN202511810387.1. This solution only collects basic electrical parameters such as voltage and current, and relies on fixed thresholds to achieve fault alarms. It suffers from the problem of single-dimensional monitoring data, and does not integrate multi-physical field information such as contact temperature rise, circuit breaker mechanical waveform, partial discharge, and cabinet environment, making it impossible to achieve multi-dimensional data joint diagnosis. At the same time, this solution adopts a full data cloud architecture, lacks edge local computing and processing capabilities, and has a lag in transient fault response. Normal load fluctuations of equipment can easily generate a large number of invalid alarms, and the problems of missed or misjudged early hidden faults are prominent.

[0004] The second category is single-dimensional digital twin calibration schemes. For example, reference document CN122310869A discloses a digital twin-driven dynamic self-calibration system and method for switchgear temperature models. This scheme only performs twin model calibration and simulation on single-dimensional data of equipment temperature, and can only achieve temperature deviation correction. It does not integrate multi-dimensional fault characteristics such as harmonic distortion, partial discharge, and circuit breaker mechanical characteristics, and the data fusion dimension is extremely limited. Moreover, it only deploys cloud calibration logic, without edge lightweight simulation and local fault handling functions, and cannot achieve local rapid fault isolation, resulting in extremely poor real-time operation and maintenance.

[0005] The third category is edge fault isolation and monitoring solutions. A representative prior art document, CN121461597B, discloses a method for fault detection and isolation of intelligent power distribution systems based on edge computing. This solution relies on edge computing to achieve simple fault identification and circuit disconnection and isolation, but it only judges faults based on a single electrical parameter and lacks a multi-physics coupling analysis mechanism. At the same time, it does not set up hierarchical early warning and handling logic, and there is no differentiated handling for minor equipment degradation and serious faults, which can easily cause unnecessary power outages. Furthermore, it does not have the capabilities of digital twin self-calibration, AI fault tracing, and closed-loop iteration of work orders throughout the entire life cycle, resulting in a very low level of intelligence.

[0006] The fourth category is conventional AI diagnostics and full-cycle data management solutions, represented by document CN202510474066 and existing commercial OPC UA operation and maintenance platforms. These solutions can achieve basic multimodal data acquisition, LSTM feature extraction, and full-cycle data storage for equipment, but they generally have technical shortcomings: data fusion uses fixed weight configuration, which cannot adapt to the dynamic adjustment of equipment under light and heavy load conditions; there is no standardized fault map matching threshold and unknown fault identification mechanism; the digital twin model is a static and fixed model, without an automatic gradient descent self-calibration process, and cannot compensate for sensor drift and equipment aging deviations; operation and maintenance work orders only support one-way dispatch, and there is no closed-loop optimization mechanism for the reverse iteration model and knowledge graph of inspection and repair data.

[0007] In addition, existing industry-standard solutions generally suffer from fragmented communication protocols and disjointed operation and maintenance processes. Different manufacturers' sensing devices and main control units of power distribution cabinets have inconsistent protocols, which can easily lead to data silos. Alarm, maintenance, archiving, and model optimization are all independent of each other and rely on manual intervention, making it impossible to achieve automated closed-loop operation and maintenance. At the same time, the lack of a unique digital identity for equipment makes it impossible to achieve full lifecycle data traceability and digital accumulation of operation and maintenance experience.

[0008] In summary, all existing technical solutions are fragmented implementations of single functional modules. No single technology can simultaneously achieve multi-physics adaptive weighted fusion diagnosis, edge millisecond-level local hierarchical interlocking protection, digital twin timed automatic self-calibration, standardized map for accurate traceability, and closed-loop iterative optimization of operation and maintenance data. The industry urgently needs an integrated, adaptive, highly accurate, and fully closed-loop intelligent operation and maintenance system for the entire lifecycle of low-voltage distribution cabinets. Summary of the Invention

[0009] Purpose of the invention The core objective of this invention is to overcome the technical shortcomings of existing technologies, such as single monitoring dimensions, cloud processing delays, distortion of static twin models, low accuracy of fault diagnosis, fragmented operation and maintenance processes, and inability to connect data throughout the entire lifecycle of equipment. It provides a multi-physics field coupled cloud-edge collaborative intelligent operation and maintenance system for low-voltage distribution cabinets throughout their entire lifecycle. This system achieves adaptive weighted fusion analysis of multi-dimensional sensor data, millisecond-level hierarchical fault isolation at the edge, automatic calibration of digital twin models, unified access of heterogeneous devices with multiple protocols, AI-based precise fault tracing, and fully automated closed-loop iteration of predictive operation and maintenance work orders. This significantly reduces the probability of missed or misjudged faults, shortens fault handling time, reduces manual inspection input, and comprehensively improves the intelligence of low-voltage distribution cabinet operation and maintenance and the reliability of power supply. Technical solution

[0010] A multi-physics field coupled cloud-edge collaborative low-voltage power distribution cabinet full life cycle intelligent operation and maintenance system is divided into four layers from top to bottom: cabinet-side multi-physics field perception layer, edge collaborative computing gateway layer, cloud-based digital twin operation and maintenance platform layer, and mobile operation and maintenance terminal layer. It includes an electrical acquisition unit, a multi-physics sensing unit, and a local execution unit;

[0011] The electrical data acquisition unit collects real-time data on voltage, current, harmonic distortion rate, three-phase imbalance, power factor, and active / reactive power of each incoming and outgoing circuit of the low-voltage distribution cabinet. The multi-physics sensing unit integrates a wireless contact temperature sensor, an ultra-high frequency partial discharge sensor, a circuit breaker opening and closing coil waveform acquisition module, a temperature and humidity sensor, a water immersion sensor, a smoke sensor, and a cabinet door infrared access control sensor, simultaneously acquiring multi-dimensional physical signals in addition to electrical signals, including mechanical, insulation, and environmental signals. The local execution unit includes an intelligent electric circuit breaker, a local audible and visual alarm, and an in-cabinet touch screen. It receives local control commands from the edge gateway and performs local alarm and circuit breaker isolation operations. The edge gateway communicates bidirectionally with the cabinet-side perception layer and the cloud platform, and has five functional units: a multi-protocol standardized semantic conversion unit, a multi-physics field data fusion and filtering unit, a local hierarchical early warning decision unit, a lightweight twin simulation unit, and a local cache storage unit.

[0012] 1. Multi-protocol standardized semantic conversion unit Built-in Modbus-RTU, Modbus-TCP, IEC61850, LoRa, and 4G / 5G protocol parsing libraries unify the parsing of heterogeneous raw data from different brands of sensor terminals and low-voltage distribution cabinet controllers, converting it into OPC UA standardized semantic data and eliminating data silos across devices.

[0013] 2. Multiphysics Data Fusion Filtering Unit Built-in three-level processing logic: time alignment, joint noise reduction, and adaptive weighted coupling feature construction. (1) Timing alignment: unify the sampling timestamps for electrical, temperature, partial discharge, and mechanical waveform acquisition to eliminate time differences in multi-sensor acquisition; (2) Joint noise reduction: Kalman filter is used to filter out sensor zero-point drift noise, and wavelet layer decomposition is used to remove power frequency interference and instantaneous pulse interference; (3) Adaptive weighted coupling feature construction: Taking 80% of the rated load of the equipment as the dividing line, the load ≥80% is judged as a heavy load condition, and the weight of temperature and partial discharge is increased; the load <80% is judged as a light load condition, and the weight of harmonics and mechanical waveforms is increased. The weights of the four types of sensor data are dynamically allocated according to the real-time load and cumulative running time of the equipment. After weighted fusion, a unified "temperature rise-harmonic-partial discharge pulse-mechanical coil waveform" coupled fault feature vector is generated.

[0014] 3. Local tiered early warning decision-making unit Two levels of independent processing logic are configured to keep local computation latency within 100ms: Level 1 warning (minor equipment degradation risk): Alarm records are only stored locally, maintenance reminders are pushed to the cloud, no tripping command is output, and normal power supply is maintained; Level 2 warning (short circuit, severe insulation breakdown, contact burning emergency fault): Immediately issue a circuit breaker tripping command to isolate the faulty circuit, and at the same time send an interlocking protection signal to the upstream high-voltage power distribution equipment through the IEC60870-5 protocol to prevent the fault from spreading upstream.

[0015] 4. Lightweight twin simulation unit Deploy a lightweight digital sub-model of the equipment, synchronize cabinet-side sensing data in real time, simulate equipment degradation trends locally for a short period of time, output a preliminary health score, and upload it to the cloud to complete a global accurate assessment.

[0016] 5. Local cache storage unit The system continuously stores at least 7 days of complete original sensor data. When the low-voltage power distribution outdoor network is interrupted, the system maintains offline monitoring, local early warning and fault isolation functions. After the network is restored, the missing data is resumed according to the time breakpoint to ensure data integrity. It includes a global digital twin modeling and self-calibration unit, an LSTM fault root cause tracing unit, a full lifecycle data management unit, a predictive maintenance work order closed-loop unit, and a fault knowledge graph library unit.

[0017] 1. Global Digital Twin Modeling and Self-calibration Unit Assign a globally unique digital identity ID to each low-voltage distribution cabinet and build an initial static twin model based on the equipment's factory 3D parameters; Self-calibration execution logic: The calibration process is automatically triggered during low-load periods of the equipment each day. 24-hour continuous operation sensor data is collected as samples. A loss function is constructed based on the theoretical standard operating curve of the components. The gradient descent algorithm is used to iteratively correct the internal parameters of the twin model and compensate for model deviations caused by sensor drift and aging of metal contacts. The remaining service life of each core component is calculated based on the simulation curve after calibration.

[0018] 2. LSTM Fault Root Cause Analysis Unit The adaptive weighted multiphysics coupling feature vector uploaded from the edge gateway is used as the input to the LSTM network, and the network outputs the fault type and the precise location of the faulty cabinet circuit. The output feature vector is then matched with standard samples from the fault knowledge graph, with a similarity threshold of 85% set. If the similarity is ≥85%, the corresponding fault type is directly locked; if the similarity is <85%, it is determined to be a new unknown fault and is automatically stored in the fault sample library to be labeled. The knowledge graph will be updated after manual labeling; when multiple fault features are superimposed, the primary fault root cause is output according to the fault hazard level weight and the corresponding standard handling plan is matched.

[0019] 3. Full Lifecycle Data Management Unit Using the device's unique digital identity ID as an index, the system links the entire process of data, including the device's factory parameters, commissioning and acceptance records, regular inspection records, fault repair records, component replacement ledgers, and scrapping archives, forming a completely traceable data chain.

[0020] 4. Predictive Maintenance Work Order Closed-Loop Unit Based on equipment health scores and remaining component lifespan, planned maintenance work orders are automatically generated, and intelligent work order dispatch is completed in conjunction with the daily load of each maintenance team. After maintenance personnel complete on-site maintenance, they upload maintenance photos and retest operation data. After the data is uploaded, the fault knowledge graph is updated in real time, and the local simulation parameters of the cloud twin model are optimized in real time, forming a complete closed-loop linkage logic of "early warning alarm - work order dispatch - on-site maintenance - model iteration".

[0021] 5. Fault Knowledge Graph Base Unit It stores typical fault coupling characteristics, standard maintenance procedures, and spare parts models for various low-voltage switchgear; new fault samples and maintenance plans are automatically updated in the diagram, realizing the digital accumulation of operation and maintenance experience. It includes a mobile APP, a tablet maintenance terminal, and wireless communication with the cloud platform; functions include real-time hierarchical alarm push, 3D digital twin visualization viewing, maintenance work order reception and backfilling, remote issuance of low-voltage distribution cabinet opening and closing control commands, and historical fault record query. 1. The system adds an in-cabinet image recognition module, which links with the built-in camera in the cabinet to capture real-time images. The YOLO lightweight target detection algorithm is used to identify potential hazards such as abnormal cabinet door opening, cable damage, and foreign object accumulation inside the cabinet. The recognition results are used as new features and combined with multi-physics adaptive weighted coupling feature vectors to carry out fault diagnosis.

[0022] 2. The cloud platform has a built-in power quality optimization algorithm module that generates reactive power compensation and load balancing adjustment strategies based on full-circuit harmonics and three-phase imbalance data. These strategies are then distributed to the edge gateway for execution, simultaneously achieving safe operation and maintenance as well as energy saving and consumption reduction. Beneficial effects

[0023] 1. This invention, through multi-physics field time-series alignment, joint noise reduction, and adaptive dynamic weighted coupling feature joint analysis, abandons the rudimentary judgment mode of existing technologies with single parameters and fixed thresholds. It can effectively identify early hidden insulation and contact connection faults, significantly reduce invalid alarms and fault omissions, reduce the workload of maintenance personnel in manual identification, and solve the industry pain point of low fault diagnosis accuracy in existing technologies.

[0024] 2. Adopting a cloud-edge layered collaborative architecture, the edge gateway completes local data processing, fault diagnosis and local isolation, with fault handling response latency controlled within 100ms, which is far superior to traditional cloud-based centralized handling solutions; at the same time, a graded handling mechanism for minor and severe faults is set up to distinguish the risk level of equipment degradation, avoid unplanned power outages caused by minor equipment fluctuations, and significantly improve the continuity and stability of power supply.

[0025] 3. Compared with existing static twin models and single-dimensional calibration techniques, this invention has a timed automatic gradient descent self-calibration logic, which can adaptively compensate for model deviations caused by sensor drift and component aging around the clock, completely solving the problem of simulation model distortion, and significantly improving the accuracy of equipment health and remaining life simulation assessment.

[0026] 4. Set up a multi-protocol standardized semantic conversion unit, which is compatible with mainstream industrial communication protocols on the market, realizes unified and standardized access of heterogeneous equipment data, completely eliminates data silos, eliminates the need for customized modification of various sensing devices, and greatly reduces the adaptation cost of intelligent transformation of power distribution rooms.

[0027] 5. Relying on the unique digital identity of the equipment, a full lifecycle traceability data chain is built. Combined with automated work order closed loop and digital fault knowledge graph, predictive maintenance replaces traditional periodic manual inspections. This enables the digital accumulation of maintenance experience, avoids the problem of personnel experience loss, significantly reduces the frequency of on-site manual inspections, and shortens the time for fault diagnosis and handling.

[0028] 6. Configure the edge gateway with local offline caching and breakpoint resume mechanism. It can still independently complete monitoring, alarm and fault isolation in the event of external network interruption. It is suitable for use in outdoor power distribution rooms and remote factory power distribution rooms without stable external network, and the system adaptability and operation stability are greatly improved. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the four-layer overall architecture topology of the system of the present invention; 1- Cabinet-side multi-physics sensing layer; 2- Edge collaborative computing gateway layer; 3- Cloud-based digital twin operation and maintenance platform layer; 4- Mobile operation and maintenance terminal layer.

[0030] Figure 2 A block diagram of the functional modules within the edge collaborative computing gateway layer; Figure 3 A block diagram of the functional modules of the cloud-based digital twin operation and maintenance platform. Figure 4 This is a schematic diagram of the process for temporal alignment, noise reduction, and adaptive weighted coupling feature extraction of multiphysics data. Figure 5 This is a flowchart of the local hierarchical early warning and upstream and downstream interlocking fault isolation control logic of the present invention. Detailed Implementation

[0031] The following description, in conjunction with Figures 1 to 5, uses the operation and maintenance scenario of the low-voltage distribution cabinet in the 10kV distribution room of the park as a preferred embodiment to illustrate the complete implementation of the system of the present invention.

[0032] Example 1: Hardware Selection and Deployment of the Entire System Application scenario of this embodiment: power distribution room in an industrial park, equipped with 12 GGD low-voltage distribution cabinets, rated output current 630A, rated load dividing point is based on 80% rated load, i.e. 504A, to divide light and heavy load conditions.

[0033] Cabinet-side multi-physics sensing layer hardware configuration Electrical data acquisition unit: Uses Modbus-TCP multi-function power meter to acquire three-phase voltage, current, 0-31st harmonics, and three-phase unbalance. Multi-physics sensing unit: contact wireless temperature sensor (LoRa), ultra-high frequency partial discharge sensor, opening and closing coil waveform acquisition module, temperature and humidity / water immersion / smoke sensor, cabinet door infrared access control; Local execution unit: electric molded case circuit breaker, 4-20mA audible and visual alarm, 7-inch local touch screen.

[0034] Edge Collaborative Computing Gateway Hardware Selection An industrial edge gateway is selected, equipped with a quad-core industrial processor, and has built-in Modbus, IEC61850, LoRa, and 4G full network compatibility modules. It has 8GB of local storage and can store 7 days of raw sensor data in a loop. The gateway's local computing power can complete data filtering, feature fusion, and hierarchical early warning judgment within 100ms.

[0035] Cloud platform deployment The cloud platform uses cloud servers to deploy a digital twin platform, an LSTM fault tracing model, and a work order management module. The LSTM model is pre-trained offline using power distribution cabinet fault samples from the park over the past three years, and incorporates 85% similarity matching logic. The twin model's gradient descent self-calibration process is automatically triggered at 2 AM every day (during periods of low equipment load).

[0036] Mobile maintenance terminal: Android / iOS maintenance APP, supporting 4G / 5G wireless communication.

[0037] Example 2: Complete Implementation Steps (Refer to Figures 4 and 5) Step 1: Multi-source data time series alignment and adaptive weighted fusion The edge gateway standardizes the sampling period of all sensors to 100ms and uses a unified timestamp; it reads the loop current in real time. When the loop current is ≥504A (80% of rated load), the weights of temperature and partial discharge features are increased to 0.45, and the weights of harmonics and mechanical waveforms are increased to 0.05; when the current is <504A, the weights of harmonics and mechanical waveforms are increased to 0.45, and the weights of temperature and partial discharge are increased to 0.05; after Kalman filtering and wavelet noise reduction, a coupled feature vector is generated.

[0038] Step 2: Edge-based local hierarchical early warning interlocking control Example of Level 1 warning: The contact temperature rises above 45°C but does not reach 65°C, which is judged as slight deterioration. Only a maintenance work order is pushed to the cloud, and the circuit breaker does not trip. Example of a Level 2 warning: When the amplitude of the partial discharge pulse exceeds the standard and the three-phase short-circuit current exceeds three times the rated value, the gateway issues a trip command within 100ms and simultaneously sends an interlocking signal to the upstream high-voltage switchgear via the IEC60870-5 protocol to cut off the upstream power supply.

[0039] Step 3: Implementation of Daily Self-Calibration of Cloud-Based Digital Twin The system collects 24 hours of continuous operation data at 2 AM every day, constructs a loss function based on the standard temperature rise and harmonic curves of the components, iteratively corrects the simulation parameters of the twin model using gradient descent, compensates for the zero-point drift error of 0.2℃ / month of the sensor, and outputs the remaining service life of the circuit breaker and busbar.

[0040] Step 4: LSTM Fault Root Cause Analysis and Unknown Fault Handling The edge-up coupled feature vector is input into the LSTM network, which outputs the fault location; similarity calculation is performed with the standard samples in the map. Similarity ≥ 85%: Directly matches standard faults and troubleshooting solutions for "poor contact / insulation aging / mechanical jamming"; Similarity < 85%: Automatically stored in the sample library to be labeled, and automatically updated in the knowledge graph after on-site labeling by maintenance personnel.

[0041] Step 5: Closed-loop iterative implementation of maintenance work orders The system automatically generates maintenance work orders based on health scores below 70 and intelligently dispatches work orders according to shift load. After maintenance personnel perform on-site maintenance, they upload retest data, and the platform updates the fault map in real time and iterates the simulation parameters of the twin model locally to complete closed-loop self-optimization.

[0042] Example 3: Implementation of Additional Optimization Module Image recognition module: 2-megapixel high-definition camera inside the cabinet, YOLO lightweight model, detects cabinet door opening, cable damage, and foreign object accumulation at 1 frame per second, and uses recognition features and multi-physics field coupling vectors for joint diagnosis; Power quality optimization module: The cloud generates reactive power compensation switching strategies based on harmonic and three-phase imbalance data, and sends them to the edge gateway to control the low-voltage compensation capacitor bank to reduce the system harmonic distortion rate.

[0043] Example 4: Implementation effect of offline breakpoint resume download During the longest period of 48 hours when the outdoor power distribution network is disconnected, the edge gateway independently completes monitoring, early warning, fault isolation, and local storage of complete sensor data; after the network is restored, all offline data is resumed according to the timestamp breakpoint, with no data loss.

[0044] The actual operating effect of this embodiment compared to existing technologies This embodiment underwent a 6-month field trial run, resulting in a significant reduction in false alarms and missed faults; the on-site fault isolation response was stably controlled within 80-100ms; the twin model underwent daily self-calibration, and the lifetime prediction error was controlled within 5%; no manual periodic sensor calibration was required, reducing the frequency of manual inspections by 60%, and no unplanned power outages occurred.

Claims

1. A multi-physics field coupled cloud-edge collaborative low-voltage power distribution cabinet full life cycle intelligent operation and maintenance system, comprising a four-layer hierarchical architecture: cabinet-side multi-physics field perception layer, edge collaborative computing gateway layer, cloud-based digital twin operation and maintenance platform layer, and mobile operation and maintenance terminal layer; The cabinet-side multi-physics sensing layer includes an electrical acquisition unit, a multi-physics sensing unit, and a local execution unit. The electrical acquisition unit collects electrical operating parameters of each circuit of the low-voltage distribution cabinet. The multi-physics sensing unit simultaneously collects multi-dimensional physical signals of the cabinet's mechanical, insulation, and environmental aspects. The local execution unit is used to receive control commands to realize local alarm and circuit breaker operation. The mobile terminal layer communicates wirelessly with the cloud-based digital twin maintenance platform to receive alarms, view the device twin screen, process maintenance work orders, and remotely issue device control commands. Its features are, The edge collaborative computing gateway layer is communicatively connected to the cabinet-side perception layer and the cloud platform, and includes a multi-protocol standardized semantic conversion unit, a multi-physics field data fusion filtering unit, a local hierarchical early warning decision unit, a lightweight twin simulation unit, and a local cache storage unit. The multi-protocol standardized semantic conversion unit has a built-in library of multiple industrial communication protocol parsing methods, which converts the raw data of heterogeneous devices into OPC UA standard semantic data in a unified manner. The multi-physics data fusion filtering unit sequentially performs multi-sensor time alignment, Kalman filtering combined with wavelet hierarchical noise reduction, and adaptive weighted fusion with 80% of rated load as the boundary, dynamically adjusting the weights based on equipment load and running time to construct a multi-physics coupled fault feature vector. The local hierarchical early warning decision unit is configured with two levels of hierarchical handling logic, with a local calculation delay of ≤100ms. When a level two early warning occurs, the fault is isolated by local circuit breaker and an interlocking protection signal is sent to the upstream power distribution equipment through the IEC60870-5 protocol. The local cache storage unit cyclically stores at least 7 days of raw sensor data, operates independently when the external network is offline, and resumes transmission of all historical data after communication is restored. The cloud-based digital twin operation and maintenance platform layer includes a global digital twin modeling and self-calibration unit, an LSTM fault root cause tracing unit, a full lifecycle data management unit, a predictive operation and maintenance work order closed-loop unit, and a fault knowledge graph library unit. The global digital twin modeling and self-calibration unit assigns a unique digital identity ID to each low-voltage distribution cabinet. During the low-load period each day, a loss function is constructed based on the theoretical standard curve of the components. The gradient descent algorithm is used to iteratively correct the twin model parameters, compensate for sensor drift and component aging deviation, and simulate and output the remaining service life of the components. The LSTM fault root cause tracing unit takes the edge-up adaptive weighted coupling feature vector as the network input and performs similarity matching with the fault knowledge graph samples, setting an 85% matching threshold; when the similarity is less than 85%, the sample is stored in the fault sample library to be labeled; when multiple faults are superimposed, the primary root cause is output according to the hazard weight and a treatment plan is matched. The predictive maintenance work order closed-loop unit automatically generates maintenance work orders and intelligently dispatches them based on equipment health scores. After the maintenance site retest data is uploaded, the fault knowledge graph is updated synchronously in real time, and the local simulation parameters of the digital twin model are iteratively optimized in real time to form a complete maintenance closed-loop linkage logic.

2. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, The multi-protocol standardized semantic conversion unit has built-in Modbus-RTU, Modbus-TCP, IEC61850, LoRa, and 4G / 5G protocol parsing libraries.

3. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, Level 1 warnings only push maintenance reminders to the cloud and do not trigger circuit breakers; Level 2 warnings send interlocking signals to upstream power distribution equipment via the IEC60870-5 protocol to prevent the fault from spreading upstream.

4. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, The global digital twin modeling and self-calibration unit constructs a loss function based on the theoretical standard operating curve of the components and uses a gradient descent algorithm to iteratively correct the internal simulation parameters of the twin model.

5. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, The full lifecycle data management unit uses the device's unique digital identity ID as an index to associate and store data throughout the entire process of the device's manufacturing, commissioning, inspection, maintenance, spare parts replacement, and disposal.

6. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 2, characterized in that, It also includes an in-cabinet image recognition module, which links with the cabinet camera and uses the YOLO lightweight target detection algorithm to identify potential hazards such as abnormal cabinet door opening, cable damage, and foreign object accumulation inside the cabinet. The recognition results are combined with multi-physics field coupled feature vectors to carry out fault diagnosis.

7. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 2, characterized in that, The cloud-based digital twin operation and maintenance platform layer also has a built-in power quality optimization algorithm module, which generates reactive power compensation and load balancing adjustment strategies based on circuit harmonics and three-phase imbalance data, and sends them to the edge gateway for execution.

8. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, The local cache storage unit cyclically stores 7 days of complete original sensor data. When the external network is interrupted, it independently completes monitoring, early warning, and fault isolation. After the network is restored, it resumes the transmission of missing data according to the timestamp breakpoint.

9. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full life cycle intelligent operation and maintenance system according to claim 1, characterized in that, The adaptive weighted fusion is specifically as follows: with 80% of the rated load as the dividing line, the load ≥80% is considered a heavy load condition and the weight of temperature and partial discharge data is increased, while the load <80% is considered a light load condition and the weight of harmonics and circuit breaker mechanical waveform data is increased.

10. The multi-physics field coupled cloud-edge collaborative low-voltage distribution cabinet full lifecycle intelligent operation and maintenance system according to claim 1, characterized in that, The system is based on a four-layer architecture and collaboratively executes the following six steps: unified standardization and transformation of multi-source sensor data, adaptive weighted coupling feature extraction, edge local hierarchical fault handling, timed self-calibration of twin models, LSTM graph matching root cause analysis, automatic work order dispatch and model iteration closed loop.

Images