Intelligent power utilization safety management system

Abstract

The invention discloses an intelligent power utilization safety management system, and the system employs a three-level hierarchical intelligent architecture: an edge monitoring terminal which is internally provided with a lightweight CNN arc fault recognition model, achieves the millisecond-level local diagnosis and opening, and completes the local autonomy of safety risks independent of a cloud end; the regional aggregation gateway performs space-time correlation analysis on multi-terminal events through a directed graph algorithm, accurately locates a fault source and executes cooperative isolation control; and the cloud management platform fuses multi-source data, predicts a load through an LSTM model, generates a comprehensive energy efficiency strategy, completes closed-loop management and control from analysis to execution, and rapidly enables a new park by using a transfer learning technology. According to the method, the crossing from passive monitoring to active early warning, from local control to global optimization and from experience operation and maintenance to data intelligence is realized, and the safety and economical efficiency of power utilization in the park are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the technical field of comprehensive electricity management, and specifically relates to a smart electricity safety management system. Background Technology

[0002] With the deepening of smart park and smart grid construction, the complexity and requirements of park electricity management are increasing. Traditional electricity monitoring systems mainly rely on manual inspections and simple data collection, resulting in problems such as delayed response and extensive operation and maintenance. In recent years, IoT-based electricity management systems have become more widespread, but most of them adopt a two-tier architecture of data collection to cloud analysis, which has inherent defects.

[0003] The prior art (CN109767054A) discloses an energy efficiency cloud assessment method and an edge energy efficiency gateway based on deep neural networks. This scheme collects electricity consumption data through the edge gateway and uploads it to the cloud platform, where the cloud platform performs load identification and energy efficiency assessment. However, this scheme has the following limitations: 1. Safety response delay: All complex fault identification and decision-making are carried out in the cloud. From data collection and transmission to cloud analysis and command feedback, the entire link has a delay of up to several seconds, which cannot meet the millisecond-level rapid response requirements for transient safety risks such as arc faults.

[0004] 2. Ambiguous fault location: Its edge gateway is only responsible for data collection and forwarding, and lacks the ability to collaboratively analyze events of multiple nodes in the area. When a fault occurs, the system cannot accurately locate the root cause of the fault, which brings great difficulties to operation and maintenance.

[0005] 3. Disconnected energy efficiency management: The solution focuses on energy efficiency assessment and analysis, but the generated optimization suggestions are difficult to automatically and accurately send to specific equipment for implementation. It lacks closed-loop control capabilities from analysis to execution, resulting in limited energy efficiency optimization effects.

[0006] 4. Rigid architecture: The above two-layer architecture has a coarse division of functions, which cannot take into account both the high real-time requirements of local intelligence and the computationally complex global optimization, making it difficult to adapt to the diverse and high-performance power management needs within the park.

[0007] Therefore, there is an urgent need in this field for a comprehensive solution that can break through the limitations of existing architectures and achieve real-time security, precise operation and maintenance, and proactive energy efficiency. Summary of the Invention

[0008] One objective of this invention is to overcome the shortcomings of the prior art and provide a smart electricity safety management system that can achieve millisecond-level on-site handling of safety risks, precise fault location and isolation, and automatic closed-loop execution of energy efficiency strategies.

[0009] Another objective of this invention is to provide a system architecture that is hierarchical, well-defined in its division of labor, and highly efficient in its collaboration, so as to systematically resolve the contradiction between real-time performance and computational complexity in the prior art.

[0010] To achieve the above-mentioned inventive concept and objectives, the technical solution adopted in this application is as follows: A smart electricity safety management system, the core of which lies in constructing a three-level hierarchical intelligent architecture, including: Edge monitoring terminals are deployed at the end nodes of the park's power grid, such as distribution boxes and entrances to important equipment. Their core function is local autonomy. Based on a pre-set local diagnostic model, they perform real-time analysis of the collected high-frequency electrical parameters. When high-risk fault characteristics are identified, they execute first-level local control commands independently of the cloud platform, such as driving the circuit breaker to trip, thus solving the problem of delayed safety response at the root.

[0011] The regional aggregation gateway, acting as an intermediate layer, connects to multiple edge monitoring terminals within the same region. Its core function is to perform spatiotemporal correlation analysis on event information reported by multiple edge monitoring terminals. By analyzing the time series of events and the relationship between the power grid topology, it accurately locates the root cause node and propagation path of the fault, and issues second-level coordinated control commands accordingly, such as blocking or closing the circuit breaker, to achieve precise fault isolation and prevent the accident from escalating.

[0012] Cloud management platform: As the central hub of the system, its core function is global optimization. Based on multi-dimensional data such as historical electricity consumption, electricity price, environment, and production plan, it generates a set of executable energy efficiency optimization strategies and accurately sends strategy instructions to designated terminals or device controllers through regional aggregation gateways, forming a closed-loop control from analysis to execution, and proactively improving the energy efficiency of the park.

[0013] As a further improvement of the present invention, the edge monitoring terminal includes a high-frequency data acquisition module and an edge processing unit. The high-frequency data acquisition module acquires current and voltage waveforms at a sampling frequency of not less than 20kHz to meet the identification requirements of high-frequency fault characteristics such as electric arcs. The edge processing unit has a built-in lightweight convolutional neural network (CNN) electric arc fault identification model, which can perform real-time reasoning on high-frequency waveforms and accurately identify potential hazards.

[0014] As a further improvement of the present invention, the first-level local control command includes driving the local circuit breaker or solid-state switch to trip within 10 milliseconds, which far exceeds the existing cloud control scheme and is the key to achieving proactive security defense.

[0015] As a further improvement of the present invention, the specific process of the regional aggregation gateway performing spatiotemporal correlation analysis is as follows: maintaining the power grid topology model of its jurisdiction; receiving and caching event sequences reported by multiple terminals, such as those containing node ID, event type, and timestamp; and then, based on the topological relationship and event timing, using a directed graph algorithm to perform causal reasoning, thereby intelligently identifying the initial occurrence node of the fault.

[0016] As a further improvement of the present invention, the second-level cooperative control command includes sending a blocking and closing command to the upstream node identified as the root cause of the fault, so as to completely isolate the fault at the system level and ensure the normal power supply of the non-faulty area.

[0017] As a further improvement of the present invention, the cloud management platform includes an energy efficiency optimization engine, a strategy generation module, and a strategy distribution module. The energy efficiency optimization engine integrates multi-source data and uses machine learning models, such as LSTM, to predict future loads. The strategy generation module automatically generates comprehensive energy efficiency optimization strategies, including equipment group control, reactive power compensation, and energy storage scheduling, based on the prediction results and energy efficiency targets. The strategy distribution module compiles the strategies into device-recognizable instructions and distributes them for execution, forming a closed-loop management system.

[0018] As a further improvement of the present invention, the cloud management platform also includes an electricity consumption profiling module, which extracts feature vectors of electricity consumption patterns to create a profile for each energy-consuming unit, and identifies energy efficiency anomalies by comparing them with similar units, thereby achieving refined management and continuous optimization.

[0019] As a further improvement of the present invention, the electricity consumption profiling module utilizes transfer learning technology to transfer a profiling model trained in a mature park to a new park as an initial model, and fine-tunes it with a small amount of data from the new park, thereby quickly establishing an effective energy efficiency management model and solving the "cold start" problem of the new system.

[0020] As a further improvement of the present invention, the edge monitoring terminal, the regional aggregation gateway, and the cloud management platform are configured collaboratively into a complete power management closed loop. This closed loop executes sequentially: real-time diagnosis and rapid control implemented by the edge terminal, collaborative analysis and precise isolation implemented by the regional gateway, and global optimization and policy distribution implemented by the cloud platform. These three processes are interconnected and constitute the core operating logic of the system of the present invention.

[0021] As a further improvement of the present invention, the strategy generation module allows users to customize energy efficiency optimization rules through a graphical interface, such as demand response rules based on time-of-use pricing and green electricity priority rules based on real-time carbon intensity data, which greatly enhances the practicality and flexibility of the system.

[0022] Compared with the prior art, the present invention has the following outstanding substantive features and significant progress: 1. A qualitative leap in security: By enabling local autonomy at the edge terminal and tripping the circuit breaker in 10 milliseconds, the security response is improved from the second level to the millisecond level, achieving a fundamental leap from passive alarm to active defense, with drastically different technical effects.

[0023] 2. Increased operational efficiency: Through spatiotemporal correlation analysis and precise isolation of regional gateways, fault location has been changed from regional ambiguity to node-level precision, greatly shortening the time for fault diagnosis and recovery.

[0024] 3. Closed-loop economic benefits: Through the generation and closed-loop distribution of strategies on the cloud platform, energy efficiency management has been transformed from auxiliary analysis to proactive efficiency creation, which can directly reduce the park's electricity costs and carbon emissions.

[0025] 4. Innovation and non-obviousness of the architecture: The edge, region-cloud three-level collaborative architecture created by this invention solves the inherent contradiction between real-time performance and computational complexity in the existing two-layer architecture. In addition, the application of transfer learning and user-defined rules further demonstrate the high level of intelligence and practicality of the system. Attached Figure Description

[0026] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0027] Figure 1 This is a schematic diagram of the overall system architecture according to an embodiment of the present invention; Figure 2 This is a hardware structure block diagram of an edge monitoring terminal in one embodiment of the present invention; Figure 3 This is a flowchart of a regional aggregation gateway performing spatiotemporal correlation analysis in one embodiment of the present invention; Figure 4 This is a flowchart illustrating the closed-loop energy efficiency management implemented by the cloud management platform in one embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] To achieve the above-mentioned inventive concept and objectives, the technical solution adopted in this application is as follows: Reference Figures 1-4 A smart electricity safety management system, whose main inventive concept lies in constructing a three-level hierarchical intelligent architecture, including: Edge monitoring terminal: Deployed at the end nodes of the park's power grid, such as distribution boxes and entrances to important equipment, its core function is local autonomy. The pre-set local diagnostic model analyzes the collected high-frequency electrical parameters in real time, and when high-risk fault characteristics are identified, it executes the first-level local control command independently of the cloud platform, such as driving the circuit breaker to trip, thereby solving the problem of delayed safety response.

[0030] Regional aggregation gateway: As an intermediate layer connecting the upper and lower layers, it connects with multiple edge monitoring terminals in the same area. Its core function is to perform spatiotemporal correlation analysis on the event information reported by multiple edge monitoring terminals. By analyzing the time sequence of events and the relationship between the power grid topology, it can accurately locate the root node and propagation path of the fault, and issue second-level coordinated control commands such as blocking and closing the circuit breaker to achieve precise isolation of the fault and prevent the accident from escalating.

[0031] Cloud management platform: As the central hub of this system, it generates a set of executable energy efficiency optimization strategies based on multi-dimensional data, such as historical electricity consumption, electricity prices, environment, and production plans. Through regional aggregation gateways, it accurately distributes strategy instructions to designated edge monitoring terminals or device controllers, forming a closed-loop management system from analysis to execution, and proactively improving the energy efficiency of the park.

[0032] The edge monitoring terminal in this embodiment includes a high-frequency data acquisition module, an edge processing unit, and a communication and control module. The high-frequency data acquisition module acquires current and voltage waveforms at a sampling frequency of not less than 20kHz to meet the identification requirements of high-frequency fault characteristics such as electric arcs. The edge processing unit has a built-in lightweight convolutional neural network (CNN) electric arc fault identification model, which can perform real-time reasoning on high-frequency waveforms and accurately identify potential hazards.

[0033] Specifically, the edge monitoring terminal's specific composition and selection are as follows. Its core processing unit uses a microcontroller based on the ARM Cortex-M7 core, specifically the STM32H743IIT6, which has a main frequency of up to 400MHz and a double-precision FPU, providing the necessary computing power for running lightweight neural network models. The chip's built-in 2MB Flash and 1MB RAM are sufficient to store model code and runtime data.

[0034] High-frequency data acquisition module: Electrical parameter acquisition utilizes the ADE7978 dedicated high-precision metering chip. This chip supports simultaneous sampling of 3 voltage channels and 4 current channels, with a sampling rate configurable via programming; in this embodiment, it is set to 25.6kHz. The chip incorporates a high-precision 24-bit Σ-Δ ADC and a robust energy calculation engine, capable of simultaneously providing raw waveform data for arc analysis and effective parameters for metering. The current transformer selected is the HSTS016L series, with a -3dB bandwidth of 100kHz and a phase error of less than 0.5°, ensuring high-frequency signal fidelity.

[0035] The structure of the CNN model for arc fault recognition is as follows: the input layer consists of 2560 points (corresponding to 100ms of waveform data at a sampling rate of 25.6kHz), followed by three alternating convolutional and pooling layers. The convolutional kernel sizes are 16, 8, and 4, with 32, 64, and 128 kernels respectively. Finally, two fully connected layers are connected. This model was trained in the cloud using a dataset containing over 150,000 labeled data sets, covering series and parallel arcs under resistive and inductive load backgrounds, achieving an accuracy of over 99.5%. After training, it was converted into C++ code using the TensorFlow Lite Micro framework and quantized to 8-bit integer precision. The final model size is approximately 520KB, which was compiled into firmware and burned into the Flash memory of the STM32H743.

[0036] Communication and Control Module: The communication adopts a dual-mode redundancy design. The primary communication is a 4G CAT1-based wireless communication module - Quectel EC200S, and the backup communication is a LoRa wireless module - SEMTECH SX1278. The control output section uses TI's DRV8871 high-current H-bridge driver chip, which can directly drive relays or solid-state relays with a rated current of 3.5A, ensuring fast and reliable tripping operation.

[0037] In this embodiment, the first-level local control command includes driving the local circuit breaker or solid-state switch to trip within 10 milliseconds, which far exceeds the existing cloud control scheme and can achieve proactive security defense.

[0038] In this embodiment, the specific process of the regional aggregation gateway performing spatiotemporal correlation analysis is as follows: maintaining the power grid topology model of its jurisdiction; receiving and caching event sequences reported by multiple terminals, including information such as node ID, event type, and timestamp; and using a directed graph algorithm to perform causal reasoning based on topological relationships and event timing, thereby intelligently identifying the initial node where the fault occurred.

[0039] Specifically, its hardware platform uses Huawei AR651 series industrial-grade edge computing gateway, equipped with a quad-core ARM Cortex-A53 processor with a main frequency of 1.4GHz, 4GB of memory and 32GB of eMMC storage. The gateway has multiple gigabit Ethernet ports and RS485 interfaces, and supports 5G / 4G wireless access.

[0040] The spatiotemporal correlation analysis engine runs as a software service in a Docker container on the gateway. The specific implementation process of its core algorithm is as follows: Topology modeling: During system initialization, operations and maintenance personnel import or draw a single-line diagram of the region through a web interface. The engine parses it into an in-memory data structure in the form of an adjacency list, recording the parent and child node lists of each node.

[0041] Event caching: The engine subscribes to the MQTT message bus, continuously receives and caches events reported by all terminals within its jurisdiction, with the cache window being the most recent 5 minutes.

[0042] Causal reasoning: The algorithm is triggered when a new event is received. First, it searches the cache for other events that are close in time, for example, within 100ms before and after, and on the topological path related to that event node. Then, it constructs a temporary directed graph with these events as nodes, where the direction of the edges is determined by the topological relationship and the order of time, with earlier events pointing to later events. The root cause of the fault is inferred to be the node with an in-degree of 0, that is, no earlier, topologically related event points to the root node.

[0043] For example, if terminal A (the end of the line) reports an arc, and its upstream terminal B reports an overcurrent 5ms later, then in the graph constructed by the algorithm, there exists an edge A pointing to B, with an in-degree of 0, which is determined to be the root source.

[0044] In this embodiment, the second-level collaborative control command includes sending a blocking and closing command to the upstream node identified as the root cause of the fault, thereby completely isolating the fault at the system level and ensuring normal power supply to the non-faulty area.

[0045] The cloud management platform in this embodiment includes an energy efficiency optimization engine, a strategy generation module, and a strategy distribution module. The energy efficiency optimization engine integrates multi-source data and uses machine learning models, such as LSTM, to predict future loads. The strategy generation module automatically generates comprehensive energy efficiency optimization strategies, including device group control, reactive power compensation, and energy storage scheduling, based on the prediction results and energy efficiency targets. The strategy distribution module compiles the strategies into device-recognizable instructions and distributes them for execution, forming a management closed loop.

[0046] In this embodiment, the cloud management platform also includes an electricity consumption profiling module, which extracts feature vectors of electricity consumption patterns to create a profile for each energy-consuming unit, and identifies energy efficiency anomalies by comparing them with similar units, thereby achieving refined management and continuous optimization.

[0047] Specifically, the cloud management platform is built on Alibaba Cloud public cloud and adopts a microservice architecture.

[0048] Data storage: TimescaleDB is used to store all monitored time-series data; MongoDB is used to store device metadata, topology information and policy configurations; and Redis is used as a cache and message queue.

[0049] Energy Efficiency Optimization Engine: The load forecasting model employs an LSTM network with a Seq2Seq with Attention mechanism. Model input includes: total park load data for the past 7 days at a 15-minute time resolution, along with corresponding outdoor dry-bulb temperature, relative humidity, wind speed, solar radiation intensity, and date type (weekday / weekend / holiday, etc.). The model outputs load forecasts every 15 minutes for the next 24 hours. The model's mean absolute percentage error (MAE) has remained consistently below 4.2% when trained on historical data.

[0050] The electricity consumption profiling module employs an unsupervised learning algorithm. First, it extracts 15 feature vectors for each energy-consuming unit (e.g., an injection molding machine, an office floor), including daily load curves (96 points), daily load rate, daily peak-to-valley difference rate, and minimum nighttime load. Then, it uses the K-Means++ algorithm to cluster these feature vectors. The system automatically runs the clustering analysis weekly and visualizes the results. When the load curve shape or feature value of a unit consistently deviates from the center of its cluster by more than two standard deviations, the system automatically marks it as an energy efficiency anomaly and issues an alarm.

[0051] In this embodiment, the electricity profile module utilizes transfer learning technology to transfer a profile model trained in a mature park to a new park as the initial model, and then fine-tunes it using a small amount of data from the new park, thereby quickly establishing an effective energy efficiency management model and solving the cold start problem of the new system.

[0052] In a specific case, for a newly built industrial park, its cloud platform initially lacked local data. The implementation team exported a well-trained LSTM load forecasting model and a K-Means electricity consumption profiling model from the cloud platform of another similarly established and stable industrial park. These two models were imported into the new park's platform as initial models. After the new system was operational, using only three weeks of local operational data and fine-tuning techniques to update the initial models, the MAPE (Mean Average Power Estimation) for the next 24 hours rapidly decreased from an initial 15% to a usable level below 6%, and the clustering effectiveness of the electricity consumption profiling reached over 85%, significantly shortening the system's turnaround time.

[0053] The strategy generation module allows users to customize energy efficiency optimization rules through a graphical interface, such as demand response rules based on time-of-use pricing and green electricity priority rules based on real-time carbon intensity data, which greatly enhances the system's practicality and flexibility.

[0054] In this embodiment, the edge monitoring terminal, the regional aggregation gateway, and the cloud management platform are configured together to form a complete closed loop for power management.

[0055] The closed loop is executed sequentially as follows: real-time diagnosis and rapid control implemented by the edge terminal, collaborative analysis and precise isolation implemented by the regional gateway, and global optimization and policy distribution implemented by the cloud platform. These three processes are interconnected and constitute the core operating logic of the system of this invention.

[0056] Taking the parallel arcing caused by cable insulation deterioration in a data center rack as an example, the handling process of the edge terminal is explained in detail: Signal acquisition and preprocessing: The edge monitoring terminal deployed on the front cabinet of this column uses its ADE7978 chip to synchronously sample the three-phase current and voltage at a rate of 25.6kHz. After each sampling, a 100ms long data block (2560 points) is transmitted to the RAM of the STM32H743 through the SPI interface.

[0057] Real-time inference and decision-making: The STM32H743 calls a CNN model stored in Flash to perform inference on the input current waveform data. The model completes the calculation within 15ms and outputs a fault probability value of 0.94, which far exceeds the preset risk threshold of 0.85.

[0058] First-level control execution: The edge processing unit immediately sends a high-level signal to the DRV8871 driver chip via GPIO. Within 1ms, the DRV8871 drives the connected solid-state relay to disconnect the circuit. From inference completion to complete circuit disconnection, the total response time is less than 18ms, far below the national standard requirement of 100ms, achieving true proactive defense.

[0059] Information reporting: While cutting off the power, the edge monitoring terminal sends a formatted alarm message to the regional aggregation gateway via the 4G network.

[0060] Continuing from the previous example, the handling process for the regional aggregation gateway is as follows: Event reception and topology matching: The gateway's MQTT service simultaneously receives voltage sag events from RPP_01_A (faulty cabinet) and its upstream PDU_Main_B. The spatiotemporal correlation analysis engine immediately loads the electrical topology diagram of the data center.

[0061] Causal reasoning: The engine compares the timestamps and finds that the event of RPP_01_A is earlier than that of PDU_Main_B. Combining the topological relationship, the algorithm conclusively infers that RPP_01_A is the root cause of the fault, and the voltage sag of PDU_Main_B is a derivative phenomenon affected by it.

[0062] Level 2 collaborative control: The engine generates and issues two control commands: Command 1 (to PDU_Main_B) clears irrelevant alarms to avoid interfering with operational and maintenance judgments.

[0063] Command 2 (to the upper-level circuit breaker of RPP_01_A) remotely electricalally interlocks it.

[0064] Results Reporting: The gateway reports the final analysis report to the cloud platform. The entire process is completed within 10 seconds, and the maintenance personnel receive a clear fault location: RPP_01_A, the head unit, rather than a vague description of abnormal voltage in a certain area.

[0065] Practical Benefit Analysis of Cloud-based Energy Efficiency Closed-Loop Optimization and Profiling Taking the energy efficiency optimization of a typical summer day in a manufacturing park as an example: Data Fusion and Prediction: The energy efficiency optimization engine starts at 1:00 AM every day. It calls the meteorological API to obtain the temperature and humidity forecasts for the next day accurate to the hour; obtains the time-of-use electricity price for the next day from the power grid system; and obtains the production schedules of each workshop from the park's MES system. The LSTM model predicts that there will be a peak electricity consumption period of 2 hours from 2:00 PM to 4:00 PM the next day, with a peak load expected to reach 8.5MW.

[0066] Strategy Generation: The strategy generation module aims to minimize overall electricity costs and uses a mixed-integer linear programming algorithm to generate a set of collaborative optimization strategies. Air conditioning system: From 13:45 to 16:15, the set temperature of all office air conditioning units will be increased by 1°C from the standard. This is expected to reduce peak load by 350kW.

[0067] Energy storage system: The 1MWh / 500kW energy storage power station in the park is instructed to start discharging at a constant power of 500kW from 13:30 until 16:00. It is expected to reduce the peak load by 500kW.

[0068] Production flexibility: Negotiate with a discontinuous electroplating line to move its operating hours from 14:00-15:00 to 11:00-12:00. Peak load reduction is expected to be 200kW.

[0069] Strategy Issuance and Execution: The strategy issuance module compiles the above strategies into control instructions for their respective systems and issues them to the building automation system, energy storage management system, and electroplating line PLC via the OPC UA protocol at set times. After strategy execution, the actual monitored peak load for the day was 7.38MW, successfully reducing the peak load by 1.12MW, a peak reduction rate of 13.2%. It is estimated that this resulted in a daily electricity cost saving of approximately RMB 8,500.

[0070] User profiling and continuous optimization: Weekly cluster analysis using the electricity consumption profiling module revealed that the unit product power consumption in Workshop 3 was 12% higher than that in Workshops 1 and 2. Transfer learning model-assisted diagnosis (transferring energy efficiency model features from Workshop 1 to Workshop 3 data for comparative analysis) indicated an unreasonable air compressor group control strategy, exhibiting a situation where the compressors were oversized for their intended use. After the operations and maintenance team adjusted the strategy accordingly, the workshop's total weekly energy consumption decreased by 8.5%.

[0071] The above are merely preferred embodiments of the present invention and are not intended to limit 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 electricity safety management system, characterized in that, include: Multiple edge monitoring terminals are deployed at various power consumption nodes of the park's power grid to collect electrical parameters and perform real-time analysis based on a pre-set local diagnostic model. When a fault characteristic that meets the preset conditions is identified, the terminal can execute the first-level local control command independently of the cloud platform. At least one regional aggregation gateway is communicatively connected to multiple edge monitoring terminals within the same physical or logical region, for performing spatiotemporal correlation analysis on event information from multiple terminals to determine the root cause node and propagation path of the fault, and issuing a second-level collaborative control command to the affected nodes based on the analysis results. The cloud management platform communicates with the regional aggregation gateway and is used to generate an executable set of energy efficiency optimization strategies based on multi-dimensional data, and to send the strategy instructions to the designated edge monitoring terminal or controlled device through the regional aggregation gateway.

2. The intelligent electricity safety management system according to claim 1, characterized in that, The edge monitoring terminal includes: The high-frequency data acquisition module is configured to acquire current and voltage waveform data at a sampling frequency of not less than 20kHz; The edge processing unit has a built-in lightweight local fault diagnosis model library, which includes at least an arc fault identification model trained based on a convolutional neural network. The edge processing unit is configured to: input the high-frequency data into the local fault diagnosis model for real-time inference, and when the inference result reaches a preset risk threshold, generate and execute the first-level local control command locally.

3. The intelligent electricity safety management system according to claim 2, characterized in that, The first-level local control command includes driving the circuit breaker or solid-state switch directly connected to the edge monitoring terminal to trip within 10 milliseconds.

4. The intelligent electricity safety management system according to claim 1, characterized in that, The regional aggregation gateway is configured to perform the spatiotemporal correlation analysis, specifically including: maintaining a topology graph model describing the power grid connection relationship within its jurisdiction; receiving and caching event sequences reported by multiple edge monitoring terminals; and using a directed graph algorithm to perform causal reasoning based on the topology graph model and the timestamp sequence of the events to identify the initial fault occurrence node and its propagation path in the power grid.

5. The intelligent electricity safety management system according to claim 4, characterized in that, The second-level coordinated control command includes sending a blocking and closing command to the upstream node identified as the root cause of the fault.

6. The intelligent electricity safety management system according to claim 1, characterized in that, The cloud management platform includes: The energy efficiency optimization engine is configured to integrate historical electricity consumption data, real-time / predictive electricity price information, environmental data, and park data. Production planning data is used to predict load curves for specific future periods using machine learning models; The strategy generation module is configured to automatically generate a comprehensive energy efficiency optimization strategy, including equipment group control start-up and shutdown, reactive power compensation, and energy storage system charging and discharging scheduling, based on the load forecast results and preset energy efficiency targets. The strategy distribution module is configured to compile the comprehensive energy efficiency optimization strategy into a set of control instructions that the equipment can recognize, and distribute it to the corresponding execution unit through the regional aggregation gateway.

7. The intelligent electricity safety management system according to claim 6, characterized in that, The cloud management platform also includes an electricity consumption profiling module, which is configured to create an electricity consumption behavior profile for each major energy-consuming unit or area in the park, and to identify units with abnormal energy efficiency by comparing the profiles with those of similar units.

8. The intelligent electricity safety management system according to claim 7, characterized in that, The electricity consumption profiling module is further configured to: utilize transfer learning technology to transfer a mature electricity consumption profiling model trained in a source campus to a new target campus as an initial model and perform fine-tuning.

9. The intelligent electricity safety management system according to claim 1, characterized in that, The edge monitoring terminal, regional aggregation gateway, and cloud management platform are configured to execute a complete power management closed loop, which includes, in sequence: real-time diagnosis of power safety risks and first-level rapid control process executed by the edge monitoring terminal; multi-node event spatiotemporal correlation analysis and second-level collaborative control process executed by the regional aggregation gateway; and global energy efficiency analysis, optimization strategy generation, and strategy distribution process executed by the cloud management platform.

10. The intelligent electricity safety management system according to claim 6, characterized in that, The strategy generation module allows users to customize energy efficiency optimization rules through a graphical interface. These rules include demand response rules based on time-of-use pricing and green electricity priority rules based on real-time carbon intensity data.

Images