Intelligent power distribution system based on internet of things technology and real-time fault diagnosis method based on digital twinning

Abstract

This invention discloses an intelligent power distribution system based on Internet of Things (IoT) technology and a real-time fault diagnosis method based on digital twins, belonging to the interdisciplinary field of power distribution and information technology. The system adopts a three-layer collaborative architecture of cloud-edge-device, including a terminal perception layer, an edge intelligence layer, and a cloud platform application layer connected in sequence. The terminal perception layer realizes multi-dimensional holographic perception encompassing electrical, physical, environmental, and visual dimensions; the edge intelligence layer enables localized real-time decision-making and edge autonomous control; and the cloud platform application layer constructs a digital twin synchronized with the physical entity at high frequency and an AI decision-making system. This invention also discloses a real-time fault diagnosis method based on the above system, which achieves real-time diagnosis and proactive early warning of power distribution faults through data synchronization preprocessing, multi-level anomaly detection, twin visualization mapping, predictive maintenance decision-making, and closed-loop model optimization. This invention significantly improves the operation and maintenance efficiency and power supply reliability of power distribution systems, reduces line losses and operation and maintenance costs, and can be widely applied to the intelligent upgrade of various 10kV and below voltage level power distribution scenarios.

Description

Technical Field

[0001] This invention belongs to the cross-disciplinary field of power distribution technology and new-generation information technology. Specifically, it relates to an intelligent power distribution system based on Internet of Things (IoT) technology, and also to a real-time fault diagnosis method based on digital twins. It is applicable to the intelligent operation and maintenance and management of power distribution networks with voltage levels of 10kV and below, and belongs to a key branch of power distribution IoT in the field of smart grid technology. Background Technology

[0002] As the most fundamental energy form in modern society, the safety, reliability, and efficiency of electricity distribution directly determine the stable operation of social production and daily life. Currently, the vast majority of operating power distribution systems in China are still in the traditional distribution stage, centered on circuit breakers, contactors, and metering instruments. Their operation and management models suffer from systemic defects. Furthermore, existing intelligent transformation solutions have technical limitations, making it difficult to meet the stringent requirements for power quality and reliability in the digital economy era. Specific defects are as follows: Firstly, traditional power distribution systems suffer from three major pain points: Firstly, outdated operation and maintenance models lead to severely delayed fault response. Current power distribution system status monitoring relies primarily on monthly / quarterly manual inspections, which not only incurs high manpower costs and low inspection efficiency but also fails to capture sudden and transient fault signs. Fault location depends on personnel experience, and industry data shows that the average fault repair time under traditional models is generally greater than 4 hours, causing significant downtime losses for companies with continuous production. Secondly, inefficient energy management results in low economic efficiency. Traditional power distribution systems mostly only install active energy meters at the main incoming line, lacking refined metering of each circuit and key equipment. This makes it impossible to accurately grasp the flow of electricity and loss points, resulting in non-technical line loss rates generally exceeding 5%. Furthermore, the lack of real-time reactive power compensation, load regulation, and other optimization methods leads to substandard system power factors. Thirdly, passive maintenance strategies result in high rates of sudden failures. Equipment maintenance generally adopts reactive repair or rigid periodic maintenance models. Reactive repair causes unplanned production interruptions, while periodic maintenance is prone to both over-maintenance and under-maintenance. Industry statistics show that more than 60% of power distribution faults are sudden faults without warning, which poses a great threat to safe production.

[0003] Secondly, existing intelligent transformation solutions have significant technical limitations. In recent years, most circuit breakers and smart meters with added communication modules have only achieved remote data acquisition and monitoring functions. Core intelligence relies entirely on cloud-based central processing, resulting in four major problems: First, high latency and insufficient real-time performance. The edge side lacks local decision-making capabilities; all data must be uploaded to the cloud for analysis before instructions are issued, failing to meet millisecond-level real-time requirements for relay protection and islanding detection, posing serious security risks during network anomalies. Second, bandwidth pressure and single-point failure risk. Uploading massive amounts of sensor data places extremely high demands on network bandwidth, while the cloud becomes a single point of failure; failure there will paralyze the entire system. Third, the application of digital twins is superficial. Existing solutions generally have digital twin model refresh rates below 1Hz, only suitable for post-event retrospective display. The twin model is severely out of sync with the physical entity's state, unable to support real-time simulation, prediction, and decision-making. Fourth, a lack of a complete technical architecture. It cannot integrate edge perception, edge intelligence, cloud-edge collaboration, and high-fidelity digital twins, failing to achieve closed-loop intelligence from perception to cognition to action.

[0004] In summary, existing technologies cannot fundamentally solve the systemic defects of traditional power distribution systems. There is an urgent need for an intelligent power distribution system with edge autonomy, cloud-edge collaboration, and high-fidelity real-time digital twin capabilities, as well as supporting real-time fault diagnosis methods, to achieve a leapfrog development of power distribution systems from traditional passive operation and maintenance to proactive intelligent operation and maintenance. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent power distribution system based on Internet of Things (IoT) technology. This system solves the problems of lagging operation and maintenance, inefficient energy management, and passive maintenance in traditional power distribution systems. It also addresses the technical bottlenecks of existing intelligent solutions, such as insufficient real-time performance, high bandwidth pressure, superficial application of digital twins, and incomplete architecture. Another objective of this invention is to provide a real-time fault diagnosis method based on digital twins, enabling real-time diagnosis, proactive early warning, and closed-loop optimization of power distribution faults, thereby significantly improving the power supply reliability and operation and maintenance efficiency of the power distribution system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides an intelligent power distribution system based on Internet of Things (IoT) technology. The system adopts a three-layer collaborative architecture of cloud, edge, and terminal, comprising a terminal sensing layer, an edge intelligence layer, and a cloud platform application layer connected sequentially. The terminal sensing layer is used to achieve multi-dimensional holographic perception of the power distribution system's electrical, physical, environmental, and visual states, collecting multi-source heterogeneous power distribution operation data and transmitting it to the edge intelligence layer. The edge intelligence layer employs an embedded AI edge computing gateway cluster to perform localized preprocessing, real-time intelligent decision-making, and edge autonomous control on the data collected by the terminal sensing layer, and uploads the processed data to the cloud platform application layer. The cloud platform application layer is used to construct a digital twin that is highly synchronized with the physical power distribution entity, enabling intelligent operation and maintenance, predictive maintenance, and energy efficiency optimization decisions throughout the entire power distribution lifecycle based on global power distribution data.

[0007] Furthermore, the terminal sensing layer includes an electrical parameter acquisition subsystem, a status parameter monitoring subsystem, and a hybrid communication networking unit. The electrical parameter acquisition subsystem deploys 0.5S-level high-precision CT / PT distributed measurement nodes in each circuit of the power distribution system (incoming lines, outgoing lines, and important branches). Each measurement node has local data processing capabilities, used to collect real-time power quality data across all dimensions, including three-phase voltage, current, power, power factor, up to the 31st harmonic, and voltage sag / boost. The measurement nodes also have a periodic self-calibration function to ensure long-term stability of measurement accuracy. The status parameter monitoring subsystem includes SAW wireless passive temperature sensors, digital humidity sensors, MEMS vibration sensors, and industrial-grade high-definition video cameras deployed at key electrical connection points (circuit breaker contacts, busbar connections, cable joints), used to collect temperature, ambient humidity, equipment vibration data, and on-site visual image data of the power distribution equipment. The SAW wireless passive temperature sensor obtains energy through electromagnetic coupling, requiring no battery power and has a lifespan of up to 10 years. It is designed for use for over 10 years and is suitable for harsh environments with high pressure and high temperature. The hybrid communication networking unit adopts a networking architecture that combines wired industrial Ethernet, LoRaWAN wireless networking and 5G communication. Important nodes use wired industrial Ethernet, distributed sensors use LoRaWAN wireless networking, and mobile monitoring points use 5G communication to form a reliable and efficient data transmission network.

[0008] Furthermore, the embedded AI edge computing gateway of the edge intelligence layer uses the NVIDIA Jetson Xavier NX module as its core. This module provides 21 TOPS of AI inference computing power with a power consumption of only 10-15W. The gateway hardware is equipped with multiple RS-485 / Modbus interfaces, gigabit industrial Ethernet ports, LoRaWAN communication modules, and 4G / 5G wireless communication modules, possessing powerful multi-protocol access and heterogeneous networking capabilities. The gateway adopts an industrial-grade design, supports wide temperature operation from -40℃ to 85℃, and has dustproof, moisture-proof, and vibration-resistant characteristics, making it suitable for harsh industrial environments.

[0009] Furthermore, the edge intelligence layer incorporates a lightweight AI model, a local real-time decision engine, and a streaming data preprocessing unit. The lightweight AI model is a deeply optimized and pruned YOLOv5s target detection model, with a compressed size of only 3MB. It is used to analyze the video stream from the front-end camera in real time, enabling automatic recognition of instrument pointer readings, digital dial readings, switch opening / closing status, and indicator light colors, achieving an accuracy rate of over 99.5%. The local real-time decision engine incorporates a hybrid decision system based on rule engines and machine learning algorithms. This system can independently complete multiple key tasks without cloud dependency: over-limit alarms based on real-time electrical parameters (overcurrent, overvoltage, undervoltage), overheat warnings based on temperature trend analysis, and fault isolation and power restoration to non-faulty areas based on system topology (undervoltage tripping, automatic transfer switch logic), with an anomaly response time of less than 100ms. The streaming data preprocessing unit uses streaming data processing technology to filter, compress, and extract features from the raw data collected by the terminal. This data is then cached locally via a circular buffer and uploaded to the cloud platform according to a strategy, reducing network bandwidth usage by 70%. above.

[0010] Furthermore, the cloud platform application layer includes a time-series data foundation, a 3D visualization engine, a digital twin module, and an advanced AI application platform. The time-series data foundation uses an InfluxDB time-series database cluster as its core storage engine, employing a distributed storage architecture that supports horizontal scaling and can handle terabytes of sensor data daily. It is specifically optimized for storing and quickly querying massive amounts of time-stamped sensor data. It has a built-in data quality management system with functions for data integrity checks, abnormal data cleaning, and data quality assessment. The 3D visualization engine is a high-performance 3D rendering platform built on the Three.js WebGL engine, used to construct 1:1 high-precision 3D models of power distribution equipment such as substation environments, distribution cabinets, busbars, circuit breakers, and meters. It employs Level of Detail (LOD) technology to achieve efficient rendering of large-scale scenes, with a rendering frame rate of up to 60fps, supporting access from multiple platforms including web and mobile devices. The digital twin module incorporates a real-time data push protocol based on WebSocket to achieve high-frequency synchronization between the twin model and the physical power distribution entity, with a key status data refresh rate of up to 10Hz. The above supports true real-time monitoring and interaction; it realizes multi-dimensional data visualization mapping in the 3D model, which can intuitively display the real-time dynamic flow of current and voltage data, and display the temperature distribution through color gradient, realizing the visualization penetration of physical state. At the same time, it supports spatiotemporal data backtracking, which can reproduce the system operation status at any point in time; the advanced AI application platform includes a predictive maintenance unit, an energy efficiency optimization analysis unit, and a fault root cause analysis unit.

[0011] Furthermore, the predictive maintenance unit, based on historical time-series data, constructs a multimodal deep learning model integrating LSTM (Long Short-Term Memory) network and Transformer to predict the remaining useful life (RUL) of key power distribution equipment such as circuit breakers and transformers, achieving a prediction accuracy of over 85%. The energy efficiency optimization analysis unit incorporates an energy efficiency optimization model based on big data analysis. Through cluster analysis and anomaly detection algorithms, it accurately locates energy consumption anomaly circuits, identifies energy consumption anomaly points, and provides optimization strategy suggestions such as reactive power compensation and peak-valley scheduling, which can reduce the overall line loss rate of the power distribution system from over 5% to below 3%. The fault root cause analysis unit constructs a fault diagnosis system based on knowledge graphs. After a fault occurs, it can automatically trace back multidimensional data (electrical parameters, temperature, video recordings) of the entire process before and after the fault, and assists maintenance personnel in quickly locating the root cause of the fault through causal reasoning algorithms, improving diagnostic efficiency by more than 5 times.

[0012] Furthermore, the system also includes a security protection system, which comprises a defense-in-depth network security architecture, a redundant reliability design, and a data privacy protection unit. The defense-in-depth network security architecture includes multiple security measures such as network boundary protection, communication encryption, and access control. It employs national cryptographic algorithms for end-to-end encryption of data transmission and establishes a digital certificate-based identity authentication mechanism to ensure system network security. The redundant reliability design utilizes a dual-machine hot standby architecture for critical nodes, and the edge gateway has the ability to resume transmission after network outages, allowing it to operate independently even during network interruptions. A built-in system health monitoring mechanism monitors the operating status of each component in real time, enabling self-healing from faults. The data privacy protection unit follows the principle of data minimization, anonymizes sensitive data, and establishes a data hierarchical authorization mechanism to ensure data security and controllability, complying with GDPR, the Cybersecurity Law, and other relevant regulations.

[0013] On the other hand, the present invention provides a real-time fault diagnosis method based on digital twins, the method being implemented based on the aforementioned intelligent power distribution system, and comprising the following steps: S1. Data Synchronization and Preprocessing: The edge intelligence layer collects multi-source sensor data from the terminal perception layer at a frequency of no less than 10Hz, and uploads it to the cloud platform application layer with low latency via the MQTT protocol. An adaptive data compression algorithm is designed to dynamically adjust the transmission strategy according to network conditions. The cloud platform receives and stores data in real time through a time-series database, while driving the real-time update of the 3D twin model. A data quality monitoring mechanism is established to ensure the integrity, accuracy, and timeliness of the data. S2. Multi-level Anomaly Detection: Establish an edge-cloud collaborative anomaly detection system. Lightweight AI models and rule engines are deployed at the edge for the first level of anomaly screening, including transient anomalies such as sudden current surges, with a response time of less than 100ms. Deep AI models are deployed in the cloud to perform deep scanning of global data, using unsupervised learning algorithms to identify more complex and latent anomaly patterns, including gradual anomalies such as slowly increasing harmonic content and exacerbated three-phase imbalance. An anomaly scoring system is established to classify the severity of detected anomalies. S3. Intelligent Diagnosis and Visualization Mapping: Construct a diagnostic mapping mechanism based on digital twins. Once an anomaly is detected, the system immediately locates and visualizes the anomaly point in the digital twin: if the outgoing circuit current exceeds the limit, the circuit will be highlighted and flashing in the 3D model and an alarm will be triggered; if the cable joint temperature is abnormal, the joint will turn red in the model and display the precise temperature value; at the same time, a fault propagation model is established to predict the possible scope and development trend of the anomaly. S4. Predictive Maintenance and Decision Support: Integrate predictive maintenance models to assess the probability and time of failure if the anomaly continues to develop; automatically generate security isolation schemes and load transfer schemes based on the system topology and push them to maintenance personnel in the form of work orders; establish a decision support knowledge base to provide handling suggestions and emergency plans to assist maintenance personnel in making optimal decisions. S5. Closed-loop verification and model optimization: After maintenance personnel complete on-site operations or repairs according to platform instructions, the repair results and subsequent equipment status data are fed back to the system to verify the accuracy of the diagnosis and optimize the AI ​​model; a continuous learning mechanism is established to continuously optimize the diagnostic rules and model parameters through reinforcement learning algorithms, forming a continuously self-improving intelligent closed loop.

[0014] Compared with the prior art, the present invention has the following significant advantages: The invention achieves significant technical results and overcomes existing technological bottlenecks. It constructs a complete three-layer collaborative technical architecture of cloud-edge-device, realizing closed-loop intelligence from perception to cognition to action. This overcomes the technical bottlenecks of insufficient multi-source data fusion, lack of edge intelligence, low fidelity of digital twins, and imperfect cloud-edge collaboration in existing technologies. Through multi-dimensional holographic perception, it achieves accurate acquisition of the entire state of the power distribution system; through edge autonomy and cloud-edge collaboration, it balances millisecond-level real-time control with global intelligent optimization, reducing network bandwidth usage by more than 70% and solving the problems of high latency and single point of failure in existing solutions; through a high refresh rate digital twin, it achieves high-frequency synchronization of physical entities and digital models at over 10Hz, with the model refresh rate more than 10 times higher than existing solutions, solving the pain point that traditional digital twins can only display information after the fact and cannot make real-time decisions; through an AI algorithm system, it achieves true predictive maintenance, with an equipment fault warning accuracy rate of over 85%, fundamentally preventing the vast majority of sudden power outages.

[0015] With outstanding economic benefits, this invention significantly reduces the total lifecycle cost. It can shorten the average fault repair time from 4 hours to less than 15 minutes, achieve a fault location accuracy of 99%, and reduce production interruption losses by more than 90%. Through refined energy efficiency management, it can reduce line loss rate by 30-50%, significantly saving electricity costs. Through predictive maintenance, it can extend the service life of major power distribution equipment by 20-30% and reduce fault repair costs by 60%. At the same time, it can reduce manual inspection costs by more than 80%, significantly reducing the investment in operation and maintenance manpower. The project has a short investment payback period and has extremely high economic value.

[0016] With broad social benefits and strong promotional value, this invention can significantly improve the power supply reliability of power distribution systems, providing higher quality power supply for industrial production, commercial operations, and residential electricity use; reduce carbon emissions through energy efficiency optimization, contributing to the achievement of dual-carbon goals; and not only provides a complete system architecture, but also a complete methodology for the intelligent upgrading of power distribution systems. It can be widely applied to various power distribution scenarios such as industrial manufacturing parks, commercial complexes, data centers, hospitals and schools, rail transit, new energy microgrids, and electric vehicle charging networks, providing key distribution-side technical support for the construction of new power systems and promoting the digital and intelligent upgrading of the power distribution industry. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0018] This embodiment takes the intelligent transformation of the low-voltage power distribution system in a workshop of a large mining enterprise as the specific implementation scenario. The transformation object is the low-voltage power distribution system in the ore dressing workshop. The original system had 31 low-voltage distribution cabinets, covering the power supply circuits of 22 important process equipment. Before the transformation, the system line loss rate was 6.2%, and the average fault repair time was more than 4 hours. There were problems such as low efficiency of manual inspection, slow fault response, and frequent sudden faults.

[0019] I. Specific Implementation and Deployment of Intelligent Power Distribution System The terminal sensing layer hardware deployment completed on-site surveys and solution design, including power distribution room environmental surveys, equipment list statistics, load characteristic analysis, and network environment assessment. Sensor placement schemes, network topology design, and installation plans were formulated. 0.5S-class open-type CT sensors and voltage sampling lines were installed on 31 low-voltage outgoing circuits of the main incoming cabinet and 22 process equipment distribution cabinets, using non-contact installation to avoid affecting existing equipment operation. Each measurement node is equipped with a local data processing unit to collect real-time, multi-dimensional power quality data and set a periodic self-calibration cycle. SAW wireless passive temperature sensors were installed at 12 high-current circuit breaker contacts and 36 busbar connection bolts, with a normal acquisition cycle of 1 minute / time, accelerating to 1 second / time when temperatures are abnormal. Digital humidity sensors and MEMS vibration sensors were also deployed, and 2-megapixel high-definition network cameras with infrared illumination were installed in 3 core distribution cabinets. The communication network uses wired industrial Ethernet to connect core equipment and gateways, distributed sensors use LoRaWAN wireless networking, and mobile terminals use 5G communication to build a hybrid communication network.

[0020] The edge intelligence layer is configured and implemented in each power distribution cabinet, with an edge computing gateway based on an NVIDIA Jetson Xavier NX module mounted on a DIN rail, adapting to the wide temperature and dust-proof environment requirements of industrial sites. The gateway collects instrument and sensor data via RS-485 bus, supporting industrial protocols such as Modbus-RTU and DL / T645. A lightweight YOLOv5s model accelerated by TensorRT is deployed to achieve automatic identification of instrument readings and switch status, with a measured identification accuracy of 99.2%. A local real-time decision engine is configured to build a hybrid decision system, incorporating over 20 localized business rules to achieve local autonomous control without cloud dependency, with a measured anomaly response time of less than 80ms. A streaming data preprocessing unit is configured to filter, compress, and extract features from the raw data, resulting in a measured reduction of network bandwidth usage by 72%.

[0021] The cloud platform application layer deployment and digital twin construction are carried out on local servers. Cloud platform software is deployed using Kubernetes container orchestration and management, an InfluxDB time-series database cluster is created, and data archiving strategies and a data quality management system are configured. Based on CAD drawings and on-site measured data, a 1:1 3D model of the power distribution system is built using Three.js, optimizing the model's polygon count to within 2 million polygons, achieving a rendering frame rate of 60fps, and supporting multi-terminal access. A digital twin module is built, using the WebSocket protocol to achieve real-time data push, with a critical status refresh rate of 12Hz, enabling multi-dimensional data visualization mapping and spatiotemporal backtracking. An advanced AI application platform is deployed: the predictive maintenance unit builds a multimodal deep learning model based on historical data, achieving a measured remaining life prediction accuracy of 86%; the energy efficiency optimization analysis unit identifies three high-energy-consumption abnormal circuits, reducing the line loss rate to 2.8% after optimization; and the fault root cause analysis unit constructs a power distribution fault knowledge graph, improving diagnostic efficiency by more than 5 times.

[0022] The security protection system deployment adopts the national cryptographic SM4 algorithm to achieve end-to-end data encryption and establishes a digital certificate identity authentication and hierarchical access control mechanism; key nodes adopt dual-machine hot standby, and the edge gateway is configured with network outage resume function, which can run independently locally for no less than 30 days when the network is interrupted; a system health monitoring mechanism is established to achieve fault self-healing; sensitive data is anonymized to comply with the requirements of the Cybersecurity Law.

[0023] II. Specific Implementation of Real-Time Fault Diagnosis Method Based on Digital Twin S1. Data Synchronization and Preprocessing: The edge gateway collects multi-source data at a frequency of 12Hz and uploads it to the cloud platform via the MQTT protocol. The compression strategy is dynamically adjusted according to the network conditions. The cloud platform stores the data in real time and drives the digital twin to update synchronously. A data quality monitoring mechanism is established to remove invalid data.

[0024] S2, Multi-level Anomaly Detection: The first anomaly screening is completed at the edge, with a transient anomaly response time of less than 80ms, triggering local alarms and controls; the cloud continuously scans global data through a deep AI model to identify hidden gradual anomalies such as harmonic rise, and establishes an anomaly scoring system of 0-100 points, divided into three risk levels: low, medium, and high.

[0025] S3, Intelligent Diagnosis and Visual Mapping: After an anomaly is detected, the abnormal point is located and visualized in the digital twin. At the same time, a fault propagation model is built to predict the scope of the anomaly's impact and warn of the risk of secondary faults.

[0026] S4. Predictive Maintenance and Decision Support: By assessing the probability and timing of failures through predictive maintenance models, it automatically generates security isolation and load transfer solutions, generates maintenance work orders and pushes them to maintenance personnel, and simultaneously matches handling steps, emergency plans and historical cases.

[0027] S5. Closed-loop verification and model optimization: After the operation and maintenance is completed, the maintenance records and equipment operation data are uploaded. The system verifies the accuracy of the diagnosis and iteratively optimizes the model and diagnostic rules through reinforcement learning algorithms to form a self-optimizing closed loop.

[0028] III. Implementation Results After a 6-month trial run, this embodiment reduced the average fault repair time from 4 hours to 12 minutes, achieved a fault location accuracy of 99.5%, and decreased the sudden failure rate by 83%. The line loss rate decreased from 6.2% to 2.8%, saving approximately 450,000 yuan in electricity costs annually. Unmanned inspection was achieved, saving 100% of inspection manpower costs. The total investment of the project was approximately 950,000 yuan, with an estimated payback period of 2.1 years and an internal rate of return of 48%.

[0029] The above description is only a preferred embodiment of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart power distribution system based on Internet of Things (IoT) technology, characterized in that, The system adopts a three-layer collaborative architecture of cloud, edge, and terminal, including a terminal perception layer, an edge intelligence layer, and a cloud platform application layer that are connected in sequence. The terminal perception layer is used to realize multi-dimensional holographic perception of the power distribution system's electrical, physical, environmental, and visual states, collect multi-source heterogeneous power distribution operation data, and transmit it to the edge intelligence layer. The edge intelligence layer uses an embedded AI edge computing gateway cluster to perform local preprocessing, real-time intelligent decision-making, and edge autonomous control on the data collected by the terminal perception layer, and uploads the processed data to the cloud platform application layer. The cloud platform application layer is used to construct a digital twin that is highly synchronized with the physical power distribution entity, and realize intelligent operation and maintenance, predictive maintenance, and energy efficiency optimization decisions throughout the entire power distribution lifecycle based on global power distribution data.

2. The intelligent power distribution system based on Internet of Things technology according to claim 1, characterized in that, The terminal sensing layer includes an electrical parameter acquisition subsystem, a status parameter monitoring subsystem, and a hybrid communication networking unit. The electrical parameter acquisition subsystem deploys 0.5S-level high-precision CT / PT distributed measurement nodes in each circuit of the power distribution system. Each measurement node has local data processing capabilities, used to collect real-time power quality data across all dimensions, including three-phase voltage, current, power, power factor, up to the 31st harmonic, and voltage sag / boost. The measurement nodes also have a periodic self-calibration function. The status parameter monitoring subsystem includes SAW wireless passive temperature sensors, digital humidity sensors, MEMS vibration sensors, and industrial-grade high-definition video cameras deployed at key electrical connection points, used to collect temperature, ambient humidity, equipment vibration data, and on-site visual image data of the power distribution equipment. The hybrid communication networking unit adopts a networking architecture combining wired industrial Ethernet, LoRaWAN wireless networking, and 5G communication. Important nodes use wired industrial Ethernet, distributed sensors use LoRaWAN wireless networking, and mobile monitoring points use 5G communication.

3. The intelligent power distribution system based on Internet of Things technology according to claim 1, characterized in that, The embedded AI edge computing gateway of the edge intelligence layer uses the NVIDIA Jetson Xavier NX module as its core, and is equipped with multiple RS-485 / Modbus interfaces, gigabit industrial Ethernet ports, LoRaWAN communication modules and 4G / 5G wireless communication modules. It adopts an industrial-grade design and supports wide operating temperature range of -40℃ to 85℃.

4. The intelligent power distribution system based on Internet of Things technology according to claim 3, characterized in that, The edge intelligence layer incorporates a lightweight AI model, a local real-time decision engine, and a streaming data preprocessing unit. The lightweight AI model is a deeply optimized, pruned, and quantized YOLOv5s target detection model, compressed to 3MB, used for real-time analysis of on-site video streams to automatically identify instrument readings, switch opening / closing status, and indicator light colors, with an accuracy rate of no less than 99.5%. The local real-time decision engine incorporates a hybrid decision system based on rule engines and machine learning algorithms, enabling independent local autonomous control of electrical parameter over-limit alarms, overheat warnings, fault isolation, and power restoration to non-faulty areas without cloud dependency, with an anomaly response time of less than 100ms. The streaming data preprocessing unit uses streaming data processing technology to filter, compress, and extract features from raw data, locally caching data through a circular buffer and uploading it according to a strategy, reducing network bandwidth usage by more than 70%.

5. The intelligent power distribution system based on Internet of Things technology according to claim 1, characterized in that, The cloud platform application layer includes a time-series data foundation, a 3D visualization engine, a digital twin module, and an advanced AI application platform. The time-series data foundation uses an InfluxDB time-series database cluster, a distributed and scalable architecture, to store timestamped power distribution sensor data and includes a built-in data quality management system. The 3D visualization engine, built on the Three.js WebGL engine, employs Level of Detail (LOD) technology to construct 1:1 3D models of all equipment in the power distribution system, achieving a rendering frame rate of 60fps and supporting multi-platform access. The digital twin module incorporates a WebSocket-based real-time data push protocol, enabling high-frequency synchronization between the twin model and the physical power distribution entity, with a key status data refresh rate of no less than 10Hz, supporting multi-dimensional data visualization mapping and spatiotemporal data backtracking. The advanced AI application platform includes a predictive maintenance unit, an energy efficiency optimization analysis unit, and a fault root cause analysis unit.

6. The intelligent power distribution system based on Internet of Things technology according to claim 5, characterized in that, The predictive maintenance unit constructs a multimodal deep learning model that integrates LSTM and Transformer based on historical time-series data to predict the remaining service life of key power distribution equipment, with a prediction accuracy of no less than 85%. The energy efficiency optimization and analysis unit has a built-in energy efficiency optimization model based on big data analysis. It locates abnormal energy consumption loops through cluster analysis and anomaly detection, and generates optimization strategies for reactive power compensation and peak-valley scheduling to reduce the line loss rate of the power distribution system to below 3%. The fault root cause analysis unit constructs a fault diagnosis system based on knowledge graphs, which is used to automatically backtrack multi-dimensional data throughout the entire process after a fault occurs, locate the root cause of the fault through causal reasoning, and improve the diagnosis efficiency by more than 5 times.

7. The intelligent power distribution system based on Internet of Things technology according to claim 1, characterized in that, The system also includes a security protection system, which comprises a defense-in-depth network security architecture, a redundant reliability design, and a data privacy protection unit. The defense-in-depth network security architecture includes network boundary protection, communication encryption, and access control, employing national cryptographic algorithms for end-to-end data encryption and establishing an identity authentication mechanism based on digital certificates. The redundant reliability design utilizes a dual-machine hot standby architecture for critical nodes, and the edge gateway possesses the ability to resume transmission after network outages and self-heal from faults, with a built-in real-time system health monitoring module. The data privacy protection unit follows the principle of data minimization to anonymize sensitive data, establishes a data hierarchical authorization mechanism, and complies with GDPR and cybersecurity laws.

8. A real-time fault diagnosis method based on digital twins, characterized in that, The method is implemented based on the intelligent power distribution system based on Internet of Things technology as described in any one of claims 1-7, and includes the following steps: S1. Data Synchronization and Preprocessing: The edge intelligence layer collects multi-source sensor data from the terminal perception layer at a frequency of no less than 10Hz, and uploads it to the cloud platform application layer via the MQTT protocol. The cloud platform stores the data through a time-series database and synchronously drives the digital twin model to update in real time, while establishing a data quality monitoring mechanism. S2. Multi-level anomaly detection: Establish an edge-cloud collaborative anomaly detection system. The edge side completes the first anomaly screening through a lightweight AI model and rule engine, while the cloud side performs a deep scan of global data through a deep AI model to identify hidden anomaly patterns and establish an anomaly scoring system to classify the severity of anomalies. S3. Intelligent Diagnosis and Visualization Mapping: Construct a diagnostic mapping mechanism based on digital twins. After an anomaly is detected, the anomaly point is located and visualized in the digital twin. At the same time, a fault propagation model is constructed to predict the scope and development trend of the anomaly. S4. Predictive Maintenance and Decision Support: The predictive maintenance model assesses the probability and time of anomalies developing into failures, automatically generates safety isolation schemes and load transfer schemes based on the power distribution system topology, generates and pushes maintenance work orders, and provides handling suggestions and emergency plans based on the knowledge base. S5. Closed-loop verification and model optimization: After the operation and maintenance is completed, the equipment status data is fed back to the system to verify the accuracy of the diagnosis. The diagnostic rules and AI model parameters are continuously optimized through reinforcement learning algorithms to form a self-optimizing intelligent closed loop.

9. The real-time fault diagnosis method based on digital twins according to claim 8, characterized in that, In step S2, the response time of the first anomaly screening on the edge side is less than 100ms. The cloud-based deep AI model uses an unsupervised learning algorithm to identify latent anomaly patterns, including gradual anomalies such as slowly increasing harmonic content and intensified three-phase imbalance.

10. The real-time fault diagnosis method based on digital twins according to claim 8, characterized in that, In step S3, the specific method of visualization mapping is as follows: the abnormal circuit is highlighted and flashed in the three-dimensional twin model and an alarm is triggered; the abnormal temperature junction displays the temperature distribution and precise value through color gradient, so as to achieve visualization penetration of the physical state.