Internet of Things type box-type substation based on online monitoring

Abstract

The invention relates to the technical field of electric power engineering, and discloses an internet of things type box-type substation based on online monitoring, which comprises a sensor network layer, an edge intelligent node layer, a self-healing communication module layer and a cloud management platform layer, the sensor network layer is used for collecting the operation state and environmental parameters of equipment in the box-type substation in real time; the edge intelligent node layer is used for processing sensor data, executing self-adaptive monitoring and local diagnosis and communicating with the cloud; the self-healing communication module layer is used for ensuring the reliability and continuity of data transmission; and the cloud management platform layer is used for receiving the data uploaded by the edge nodes and carrying out long-term analysis, model training and remote management. Through the adaptive multi-parameter fusion monitoring technology, the sampling frequency of the sensor is dynamically adjusted (for example, the temperature is increased from one time per second to two times per second when the temperature is abnormal), and in combination with a lightweight machine learning model of an edge intelligent node, equipment faults (such as insulation degradation and overheating) can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of power engineering, and particularly to an Internet of Things type box-type substation based on online monitoring. Background Art

[0002] As an important device in urban distribution networks and industrial power supply systems, box-type substations are widely used due to their compact structure, convenient installation, and stable operation. With the development of smart grids, Internet of Things technologies based on online monitoring have gradually been applied to the operation management of box-type substations. In the prior art, common monitoring schemes include manual inspections combined with handheld devices (such as infrared thermometers and partial discharge detectors), as well as automated monitoring systems based on the Internet of Things. The latter usually collects equipment operation data by deploying fixed sensors (such as temperature and voltage sensors), and uses wireless communication technologies (such as 4G, ZigBee) to transmit the data to a central server or cloud platform for analysis, so as to achieve real-time monitoring and management of the substation operation status.

[0003] However, the prior art has significant drawbacks in practical applications. First, the manual inspection method is inefficient, with a limited monitoring frequency (1-2 times per day), difficult to capture sudden failures, and relying on the experience of operators, which easily leads to missed reports (missed report rate is about 20%) or misjudgments. Second, most existing Internet of Things monitoring systems adopt a fixed parameter acquisition mode (such as once per minute), lacking the ability of dynamic adjustment, unable to optimize the monitoring strategy according to the changes in equipment status, resulting in data redundancy or omission of key anomalies. In addition, the data processing of these systems usually relies on the cloud, resulting in response delays (5-10 seconds or even longer), a decrease in communication reliability (success rate <80%) when the network is unstable (such as in remote areas or high-interference environments), and even interruption, unable to meet the real-time requirements. In summary, the prior art has deficiencies in aspects such as monitoring accuracy, response speed, and environmental adaptability, and urgently needs to be improved to enhance the operation efficiency and power supply reliability of box-type substations. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides an Internet of Things type box-type substation based on online monitoring, which solves the problem that most existing Internet of Things monitoring systems adopt a fixed parameter acquisition mode (such as once per minute) and lack the ability of dynamic adjustment.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An Internet of Things type box-type substation based on online monitoring, the box-type substation includes a sensor network layer, an edge intelligent node layer, a self-healing communication module layer, and a cloud management platform layer; The sensor network layer is used to collect the operation status and environmental parameters of the equipment in the box-type substation in real time; The edge intelligent node layer is used to process sensor data, perform adaptive monitoring and local diagnosis, and communicate with the cloud; The self-healing communication module layer is used to ensure the reliability and continuity of data transmission; The cloud management platform layer is used to receive the data uploaded by the edge nodes, and perform long-term analysis, model training and remote management.

[0006] Preferably, the sensor network layer includes a variety of sensors, and the sensors are selected from one or more of a temperature sensor, a partial discharge sensor, a vibration sensor, a gas sensor, and an environmental sensor.

[0007] Preferably, the sensors are connected to the edge intelligent node layer by wired or wireless means, and some sensors support the sleep mode to reduce power consumption.

[0008] Preferably, the edge intelligent node layer includes an embedded processor, a storage unit, a communication interface, and a power management module; The embedded processor is used to run the adaptive monitoring module and the local diagnosis module; The storage unit is used to store firmware, local data, and machine learning models; The communication interface supports a variety of communication protocols, including LoRa, NB-IoT, ZigBee, and Ethernet; The power management module supports energy harvesting and low-power operation.

[0009] Preferably, the adaptive monitoring module adjusts the sensor sampling frequency according to the device status through a dynamic sampling algorithm; The local diagnosis module uses a machine learning model to analyze the data, judge the device health status, and trigger local control instructions.

[0010] Preferably, the machine learning model is trained in the cloud and updated to the edge node through OTA over-the-air download.

[0011] Preferably, the self-healing communication module layer includes a main communication module, an auxiliary communication module, and a network monitoring module; The main communication module is used to upload data on a low-power wide area network; The auxiliary communication module is used for short-distance communication for node networking and data relaying; The network monitoring module is used to monitor the channel quality and node status in real time, and detect communication anomalies.

[0012] Preferably, the self-healing communication module layer supports channel switching and path re-planning; When the signal strength of the main channel is lower than the threshold or the packet loss rate exceeds the standard, it automatically switches to the backup channel; Build a temporary mesh network using neighboring nodes as relays to bypass faulty nodes.

[0013] Preferably, the cloud management platform layer includes a data storage module, a data analysis module, a model training module, and a user interface; The data storage module is used to store historical monitoring data; The data analysis module is used to perform trend analysis and fault prediction; The model training module is used to train and optimize machine learning models and push them to edge nodes via OTA; The user interface is used for device status monitoring, alarm receiving, and control instruction issuing.

[0014] Preferably, the working process of the box-type substation includes: S1. The sensor network layer collects data and transmits it to the edge intelligent node layer; S2. The edge intelligent node layer performs adaptive monitoring and dynamically adjusts the monitoring strategy; S3. The edge intelligent node layer conducts local diagnosis, judges the device status, and makes decisions; S4. Upload key data to the cloud through the self-healing communication module; S5. The cloud platform conducts long-term data analysis and model optimization; S6. The operation and maintenance personnel monitor the device through the cloud platform and issue instructions.

[0015] The present invention provides an Internet of Things type box-type substation based on online monitoring. It has the following beneficial effects: 1. Through the adaptive multi-parameter fusion monitoring technology, the present invention dynamically adjusts the sensor sampling frequency (such as increasing from 1 time per second to 2 times per second when the temperature is abnormal), and combines the lightweight machine learning models (such as SVM or random forest) of the edge intelligent nodes, enabling accurate prediction of device failures (such as insulation deterioration, overheating). Compared with the fixed single-parameter monitoring or manual experience judgment in the prior art, this system uses real-time data analysis and dynamic response mechanisms, significantly improving the sensitivity and accuracy of fault detection.

[0016] 2. Through the low-power edge intelligent diagnosis technology, the system realizes local data processing and decision-making on the edge nodes (such as triggering power-off within 0.5 seconds in case of anomalies), greatly reducing the dependence on cloud communication. The response time is shortened from 10 - 30 seconds in the existing technology to less than 2 seconds. This local processing ability is particularly crucial in cases of network interruption or high latency, such as in remote areas or high-interference urban environments, where the system can still operate independently and protect the equipment. Compared with traditional IoT systems that rely on cloud analysis or the several-hour response of manual inspections, this system significantly improves real-time performance and autonomy through the innovative application of edge computing, providing a new solution for the immediate protection of power equipment.

[0017] 3. The self-healing multi-path communication network of the present invention (such as NB-IoT and ZigBee switching, path re-planning) ensures the continuity of data transmission, maintaining a success rate of 92% even in an environment with 50% node failures or signal interference, far exceeding that of existing single-mode communication systems (success rate < 80%). For example, in an urban construction interference scenario, the system automatically switches channels and relays data through neighboring nodes, with the transmission delay increasing by only 15 ms. This communication mechanism breaks through the limitations of traditional IoT systems being vulnerable to the environment and has a stronger ability to adapt to complex scenarios (such as high electromagnetic interference or remote areas), providing a creative guarantee for the reliability of online monitoring.

[0018] 4. Through online monitoring and fault prediction, the system significantly reduces the need for manual inspections and the power outage losses caused by sudden failures. In the embodiment, the initial deployment cost is about 20,000 yuan, and the annual maintenance cost is 5,000 yuan. Compared with traditional manual inspections (annual average of 30,000 yuan) or fixed-parameter monitoring (annual average of 10,000 yuan), it can save about 100,000 yuan in power outage economic losses every year. Although cost reduction itself is not a brand-new concept, combined with the accurate early warning of adaptive monitoring and edge intelligence, this system elevates the maintenance efficiency to a new level, reflecting the economic benefit innovation brought by technology integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the overall system architecture diagram of an IoT-based box-type substation according to the present invention; Figure 2 is the architecture diagram of the edge intelligent node layer of an IoT-based box-type substation according to the present invention; Figure 3 is the architecture diagram of the self-healing communication module layer of an IoT-based box-type substation according to the present invention; Figure 4 is the architecture diagram of the cloud management platform layer of an IoT-based box-type substation according to the present invention; Figure 5 is the working flow diagram of an IoT-based box-type substation according to the present invention. Detailed implementation manners

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Please refer to the attached Figure 1 - attached Figure 4 , and the embodiment of the present invention provides an Internet of Things type box-type substation based on online monitoring. The box-type substation includes a sensor network layer, an edge intelligent node layer, a self-healing communication module layer, and a cloud management platform layer; The sensor network layer is used to collect the operation status and environmental parameters of the equipment in the box-type substation in real time; The edge intelligent node layer is used to process sensor data, perform adaptive monitoring and local diagnosis, and communicate with the cloud; The self-healing communication module layer is used to ensure the reliability and continuity of data transmission; The cloud management platform layer is used to receive the data uploaded by the edge nodes, perform long-term analysis, model training, and remote management.

[0022] Specifically, the box-type substation is designed as a modular structure and is applicable to the 10kV urban distribution network. The sensor network layer includes 10 sensor units, which are respectively installed at the key parts of the transformer, high-voltage switchgear, and low-voltage distribution cabinet to collect multi-dimensional data such as temperature, partial discharge, and vibration. The edge intelligent node layer uses a core board based on an ARM Cortex-M7 microcontroller (main frequency 400MHz), equipped with 64MB of flash memory and 16MB of RAM, and runs the embedded real-time operating system FreeRTOS, which is responsible for data processing and local decision-making. The self-healing communication module layer integrates two sets of communication hardware: the main module is a Semtech SX1278 LoRa chip (operating frequency band 433MHz, transmission distance up to 5km), and the auxiliary module is a TICC2530 ZigBee chip (2.4GHz frequency band, short-distance networking), which is connected to the edge node through the GPIO interface. The cloud management platform is deployed on an Alibaba Cloud ECS server (4-core 8GB configuration), runs services using Docker containerization technology, is equipped with an Nginx reverse proxy and a MySQL database, and users can access the monitoring interface developed based on Vue.js through a browser. After the system is deployed, the sensor data is collected once per second, the edge node analyzes the data once per minute and uploads the key results to the cloud, and the communication module automatically switches channels when signal interference is detected, and the entire process does not require manual intervention.

[0023] The system can accurately predict equipment failures (such as insulation deterioration and overheating) by means of adaptive multi-parameter fusion monitoring technology, dynamically adjusting the sensor sampling frequency (e.g., increasing from once per second to twice per second when the temperature is abnormal), and combining lightweight machine learning models (such as SVM or random forest) of edge intelligent nodes. Compared with the fixed single-parameter monitoring or manual experience judgment in the existing technology, this system uses real-time data analysis and dynamic response mechanism to significantly improve the sensitivity and accuracy of fault detection. For example, in the urban distribution network test, the system successfully predicted 2 insulation deterioration events within 3 months, with a missed alarm rate of 0 and a false alarm rate of less than 5%, effectively avoiding power outages. This prediction ability stems from a technological breakthrough and reflects a high degree of creativity.

[0024] The sensor network layer includes a variety of sensors, which are selected from one or more of temperature sensors, partial discharge sensors, vibration sensors, gas sensors, and environmental sensors.

[0025] Specifically, the sensor network layer consists of the following five types of sensors: (1) The temperature sensor selects the DS18B20 digital sensor of Maxim Integrated, with a resolution of 0.0625 °C, and is installed on the surface of the transformer winding and the switch cabinet contact, a total of 4; (2) The partial discharge sensor uses a customized ultra-high frequency (UHF) antenna, with a frequency band of 300 MHz - 3 GHz and a sensitivity of -60 dBm, and is arranged inside the high-voltage switch cabinet, a total of 2; (3) The vibration sensor selects the ADXL345 three-axis MEMS accelerometer of ADI, with a range of ±16 g and a resolution of 3.9 mg / LSB, and is fixed at the bottom of the transformer shell, a total of 2; (4) The gas sensor is the SF6-AH optical sensor of Alphasense, with a detection range of 0 - 2000 ppm and an accuracy of ±5%, and is installed inside the transformer sealed cabin, a total of 1; (5) The environmental sensor selects the SHT31 temperature and humidity module of Sensirion, with a temperature accuracy of ±0.2 °C and a humidity accuracy of ±2%RH, and is placed near the ventilation opening at the top of the box. The sensors are connected to the edge node through the I2C or SPI interface, and the data acquisition period is default set to once per second. During installation, each sensor uses a dustproof and waterproof housing (IP65 rating) to ensure stability in high-temperature and high-humidity environments.

[0026] The combination of multiple types of sensors realizes the all-round monitoring of the operation status of the box-type substation. Compared with single-parameter monitoring, it significantly improves the coverage rate and accuracy of fault detection, providing a rich data basis for subsequent diagnosis.

[0027] The sensors are connected to the edge intelligent node layer by wired or wireless means, and some sensors support the sleep mode to reduce power consumption.

[0028] Specifically, the temperature sensor and the environmental sensor are connected to the edge node in a wired manner through the RS485 bus (with Modbus protocol), and the cable length does not exceed 10 meters. Shielded twisted pair is used to prevent electromagnetic interference. The partial discharge sensor and the vibration sensor are wirelessly connected through a ZigBee module (based on the IEEE802.15.4 protocol, with a transmission distance of 100 meters), and the data packet size is controlled within 128 bytes. The gas sensor adopts a low-power design, is controlled by an internal STM8L single-chip microcomputer, has a power consumption of 20mW during normal operation, and the power consumption drops to 50μW in the sleep mode. By default, data is collected every 5 minutes. When the edge node detects abnormal temperature or vibration, it is switched to the per-second collection mode through a wake-up signal (triggered by a GPIO high level). The switching of the sleep mode is controlled by the edge node through a timer, and the sleep time can be adjusted through firmware configuration (range: 1 second to 30 minutes). The wireless sensor is powered by a CR2032 button battery, and the expected service life exceeds 2 years.

[0029] The connection method combining wired and wireless takes into account both reliability and flexibility. The sleep mode design significantly reduces the system power consumption and extends the service life of the sensor, especially suitable for long-term operation in remote areas.

[0030] The edge intelligent node layer includes an embedded processor, a storage unit, a communication interface, and a power management module; The embedded processor is used to run the adaptive monitoring module and the local diagnosis module; The storage unit is used to store firmware, local data, and machine learning models; The communication interface supports multiple communication protocols, including LoRa, NB-IoT, ZigBee, and Ethernet; The power management module supports energy harvesting and low-power operation.

[0031] Specifically, the edge intelligent node adopts a customized circuit board. The core is the STM32H743 microcontroller from STMicroelectronics (with a main frequency of 480 MHz and 1 MB of internal flash memory), which is paired with an external 64 MB SPI flash memory (model W25Q512) and 32 MB SDRAM (model IS42S16400) for storing firmware (about 2 MB), 30 days of local data (about 20 MB), and machine learning models (about 1 MB). The communication interfaces include: (1) LoRa module (SX1278, power consumption 10 mW, transmission rate 300 bps - 37.5 kbps); (2) NB-IoT module (Quectel BC95, frequency band B8, power consumption 200 mW); (3) ZigBee module (CC2530, power consumption 30 mW); (4) Ethernet interface (W5500, with RJ45 socket, supporting 100 Mbps). The power management module consists of the BQ25570 energy harvesting chip from TI and a 5V / 2W solar panel, paired with a 2200 μF super capacitor. After being fully charged, it can support the node to run continuously for 48 hours, and the average power consumption of the node is controlled below 50 mW. The firmware is developed based on C language and compiled using Keil uVision5, occupying a storage space of about 1.5 MB.

[0032] The hardware integration of the edge node improves the data processing ability, the multi-protocol communication interface enhances the applicability, and the energy harvesting design reduces the dependence on external power supply, ensuring the stable operation of the system in various scenarios.

[0033] The adaptive monitoring module adjusts the sensor sampling frequency according to the device status through a dynamic sampling algorithm; The local diagnosis module uses a machine learning model to analyze the data, judge the health status of the device, and trigger local control instructions.

[0034] Specifically, the adaptive monitoring module runs on edge nodes and uses a time series analysis algorithm based on a sliding window (window size 10 seconds, step size 1 second). The algorithm logic is as follows: if the transformer temperature change rate exceeds 0.5 °C / second or the vibration amplitude exceeds 2g, the sampling frequency of the partial discharge sensor will be increased from once per minute to twice per second and restored to the default value after 10 minutes; the sampling frequency adjustment is sent to the sensor through I2C bus instructions. The local diagnosis module deploys a pre-trained support vector machine (SVM) model. The input features include temperature, vibration amplitude, and partial discharge intensity (3D vector), and the output is the device status classification (normal, overheating, insulation deterioration). The model size is approximately 300KB, the training data is based on 5000 samples in the past 6 months, and the accuracy rate reaches 95%. When the diagnosis result is "abnormal", the node controls a relay (model SRD-05VDC-SL-C, driven by 5V) through the GPIO interface to trigger an audible and visual alarm (sound intensity 80dB, red LED flashing), and records the abnormal event to the local flash memory.

[0035] Adaptive monitoring improves the pertinence of data collection and reduces redundant data transmission; local diagnosis enables rapid response, reduces the power outage risk caused by faults, and improves the intelligence level of the system.

[0036] The machine learning model is trained in the cloud and updated to the edge node through OTA over-the-air download.

[0037] Specifically, the machine learning model is trained on a cloud server (equipped with NVIDIA GTX1080 GPU) using the scikit-learn library in Python. The random forest algorithm is adopted. The input features are temperature, partial discharge intensity, and vibration data, and the output is the fault probability. The training dataset contains 10,000 historical records (collected from 10 box-type substations, time span 1 year), the training time is about 30 minutes, and the generated model file size is 800KB. The model is pushed to the edge node in an OTA manner through the LoRa network. The transmission process is divided into 10 data packets (each 80KB), and the transmission time of each packet is about 20 seconds, with a total time-consuming of about 3.5 minutes. After receiving, the edge node stores the model in the flash memory and verifies the integrity through the checksum (CRC32). The update process is automatically executed during the low-load period from 2:00 to 3:00 in the early morning to ensure that it does not affect normal monitoring.

[0038] The OTA update mechanism realizes the dynamic optimization of the model, improves the diagnosis accuracy, and at the same time requires no manual intervention, reducing the maintenance cost.

[0039] The self-healing communication module layer includes a main communication module, an auxiliary communication module, and a network monitoring module; The main communication module is used for uploading data on the low-power wide area network; The auxiliary communication module is used for short - distance communication to form a network among nodes and perform data relaying; The network monitoring module is used to monitor the channel quality and node status in real - time and detect communication anomalies.

[0040] Specifically, the main communication module adopts the Quectel BC95 NB - IoT module (supporting B8 band, power consumption 180 mW, transmission rate 25.5 kbps), communicates with the edge node through the UART interface, and uploads key data (about 200 bytes) once a minute; The auxiliary communication module uses the TICC2530 ZigBee module (power consumption 30 mW, transmission distance 150 meters), supports a star - shaped network composed of up to 10 nodes, and the data relaying delay is less than 50 ms. The network monitoring module is implemented by the edge node firmware, detects the signal strength (RSSI) and packet loss rate every 10 seconds, and determines it as abnormal when the main channel RSSI is lower than - 95 dBm or the packet loss rate exceeds 20%. The hardware is installed in a waterproof metal box with an external antenna to ensure a signal coverage range of 5 km in the urban area.

[0041] The multi - mode communication design improves the flexibility of data transmission, and the network monitoring module enhances the robustness of the system, ensuring that communication can still be maintained stably in complex environments.

[0042] The self - healing communication module layer supports channel switching and path re - planning; When the signal strength of the main channel is lower than the threshold or the packet loss rate exceeds the standard, it automatically switches to the backup channel; Uses neighboring nodes as relays to construct a temporary mesh network and bypass faulty nodes.

[0043] Specifically, when the self - healing communication module detects that the NB - IoT main channel RSSI is lower than - 95 dBm, it automatically switches to the ZigBee backup channel. The switching process is controlled by the edge node and takes less than 3 seconds. After switching, the data transmission rate drops to 9.6 kbps. The path re - planning adopts the AODV (Ad - hoc On - Demand Distance Vector) routing protocol. When a certain node fails due to power outage, neighboring nodes (within 50 meters) detect the interruption of the heartbeat signal and automatically take over the data relaying task, recalculate the shortest path to the cloud (average hop count 3 hops), and it takes about 2 seconds to update the routing table. In the test, the system still maintains a 90% data transmission success rate when simulating 50% node failures.

[0044] Channel switching and path re - planning significantly improve communication reliability, reduce the risk of data loss, and are especially suitable for urban high - interference or remote weak - signal scenarios.

[0045] The cloud management platform layer includes a data storage module, a data analysis module, a model training module, and a user interface; The data storage module is used to store historical monitoring data; The data analysis module is used to conduct trend analysis and fault prediction; The model training module is used to train and optimize machine learning models and push them to edge nodes via OTA; The user interface is used for device status monitoring, alarm receiving, and control instruction issuing.

[0046] Specifically, the cloud management platform is deployed on an Alibaba Cloud server (8 cores, 16GB, Ubuntu20.04 system). The data storage module uses the InfluxDB time series database to store temperature, vibration, and partial discharge data (about 50GB) for the past two years, supporting 1000 records written per second; the data analysis module runs the ApacheSpark framework, analyzes data once a week, and generates a temperature trend curve and a fault probability report (PDF format, about 5MB); the model training module uses TensorFlow2.8 to train a random forest model, based on 10,000 samples (training time: 1 hour), and the accuracy of the optimized model is increased to 97%; the user interface is a Web application developed based on React, running in a Node.js environment, supporting real-time display of the status of 10 box-type substations (refresh interval: 5 seconds), and users can send instructions by clicking the "power off" button on the interface (delay less than 2 seconds). The platform encrypts communications through the HTTPS protocol and is equipped with a load balancer to support 100 concurrent user accesses.

[0047] The integrated analysis and remote management functions of the cloud platform improve the scalability of the system and the user experience, and the fault prediction ability reduces the incidence of power outage accidents.

[0048] Please refer to the appendix Figure 5 The working process of the box-type substation includes: S1. The sensor network layer collects data and transmits it to the edge intelligent node layer; S2. The edge intelligent node layer performs adaptive monitoring and dynamically adjusts the monitoring strategy; S3. The edge intelligent node layer conducts local diagnosis, judges the device status, and makes decisions; S4. Uploads key data to the cloud through the self-healing communication module; S5. The cloud platform conducts long-term data analysis and model optimization; S6. The operation and maintenance personnel monitor the device through the cloud platform and issue instructions.

[0049] Specifically, the temperature sensor in S1 collects data once per second (accuracy 0.1°C) and transmits it to the edge node via ZigBee (delay 10 ms); in S2, the edge node runs an adaptive algorithm. When the temperature exceeds 85°C, the sampling frequency of the partial discharge sensor is increased to 3 times per second for 15 minutes; in S3, the edge node uses an SVM model to analyze the data. If it is determined that the insulation is deteriorated (probability > 90%), the faulty circuit is shut down through a relay (action time 0.5 seconds), and the event is recorded in the flash memory; in S4, the key data (temperature 85°C, partial discharge intensity 50 pC) is uploaded to the cloud via NB-IoT (transmission time 5 seconds); in S5, the cloud analyzes the data of the past 30 days, generates a fault trend report and optimizes the model (takes 2 hours); in S6, the operation and maintenance personnel view the real-time status through the Web interface. After discovering an anomaly, an "overhaul" instruction is issued, and the instruction is transmitted to the edge node via the LoRa network (delay 3 seconds). In the simulation test of the entire process, the total time from anomaly detection to instruction execution is less than 20 seconds.

[0050] The systematic workflow realizes the full closed-loop management from monitoring to decision-making, improves the efficiency and accuracy of fault handling, and significantly reduces the need for manual intervention.

[0051] The following is an introduction in combination with specific embodiments: Embodiment 1: Practical Application of an IoT-Type Box-Type Substation Based on Adaptive Edge Intelligence Application Scenario In this embodiment, the box-type substation is deployed in a power supply system for a residential community in a 10 kV distribution network in a certain city. The community has 500 households, and the box-type substation is responsible for converting 10 kV high-voltage electricity into 400 V low-voltage electricity, with an annual power supply of approximately 2 million kWh. Due to the high temperature in summer and large electricity consumption load in this area, the transformer and switchgear often have problems such as overheating or insulation deterioration, and traditional manual inspections are difficult to meet the real-time monitoring requirements. To improve power supply reliability and reduce the risk of fault power outages, the IoT-type box-type substation designed by the present invention is adopted.

[0052] System Composition and Hardware Configuration The system of this box-type substation consists of a sensor network layer, an edge intelligence node layer, a self-healing communication module layer, and a cloud management platform layer. The specific configuration is as follows: Sensor Network Layer Temperature Sensor: The DS18B20 digital temperature sensor of Maxim Integrated is selected, with a measurement range of -55°C to +125°C, an accuracy of ±0.5°C, and is installed on the surface of the transformer winding (2 pieces) and the contacts of the high-voltage switchgear (2 pieces), and is connected to the edge node via the RS485 bus.

[0053] Partial discharge sensor: A customized ultra-high frequency (UHF) antenna is adopted, with a frequency band of 500 MHz - 1.5 GHz, a sensitivity of -65 dBm. It is installed inside the high-voltage switchgear (2 pieces), and transmits data wirelessly through the ZigBee module.

[0054] Vibration sensor: The ADXL345 triaxial accelerometer of ADI is selected, with a measurement range of ±16 g and a resolution of 3.9 mg / LSB. It is fixed at the bottom of the transformer shell (1 piece) and connected through the I2C interface.

[0055] Gas sensor: The SF6-AH optical sensor of Alphasense is selected, with a detection range of 0 - 2000 ppm and an accuracy of ±5%. It is installed inside the transformer sealed cabin (1 piece) and supports the sleep mode.

[0056] Environmental sensor: The SHT31 temperature and humidity module of Sensirion is selected, with a temperature accuracy of ±0.2 °C and a humidity accuracy of ±2%RH. It is installed at the ventilation opening on the top of the box (1 piece) and connected through the SPI interface.

[0057] All sensors are encapsulated in an IP66-rated dustproof and waterproof enclosure to ensure reliability in high-temperature and high-humidity environments.

[0058] Edge intelligent node layer Hardware: The core is the STM32H743 microcontroller of ST company (main frequency 480 MHz, 1 MB internal flash memory), with an external 64 MB SPI flash memory (W25Q512) and 32 MB SDRAM (IS42S16400).

[0059] Communication interfaces: Include LoRa module (SX1278, 433 MHz, transmission distance 5 km), NB-IoT module (BC95, B8 band), ZigBee module (CC2530, 2.4 GHz) and Ethernet interface (W5500, 100 Mbps).

[0060] Power management: The BQ25570 energy harvesting chip of TI is adopted, with a 5V / 2W solar panel and a 2200 μF super capacitor. The average power consumption is 50 mW, and it can run for 48 hours after being fully charged.

[0061] Firmware: Developed based on FreeRTOS, written in C language, occupying about 1.8 MB of storage space, and supporting adaptive monitoring and local diagnosis functions.

[0062] Self-healing communication module layer Main communication module: NB-IoT module (BC95), with a power consumption of 180 mW and a transmission rate of 25.5 kbps, connected through the UART interface.

[0063] Auxiliary Communication Module: ZigBee module (CC2530), power consumption 30mW, coverage range 150 meters, supports networking of 5 nodes.

[0064] Network Monitoring: The edge node firmware detects the signal strength (RSSI) and packet loss rate every 10 seconds. The abnormal thresholds are set as RSSI < -95dBm or packet loss rate > 20%.

[0065] The communication module is installed in a metal shielding box with an external antenna to ensure signal coverage within a 2km radius around the community.

[0066] Cloud Management Platform Layer Hardware: Deployed on Alibaba Cloud ECS server (4 cores, 8GB, Ubuntu20.04 system).

[0067] Software: Uses InfluxDB to store data (capacity 50GB), Spark for trend analysis, TensorFlow for model training, React for developing the web interface, and Nginx as a reverse proxy.

[0068] Function: Supports real-time monitoring of 10 box-type substations, processes 1000 pieces of data per second, generates weekly reports and pushes them to the user's mobile phone.

[0069] Implementation Steps and Workflow The following are the specific implementation steps of the system during actual operation: S1: The sensor network collects data and transmits it to the edge intelligent node The temperature sensor collects the transformer winding temperature once per second (such as 75°C) and transmits it to the edge node through RS485 (delay 5ms); the partial discharge sensor collects data once per minute by default (such as intensity 20pC) and transmits it through ZigBee (delay 10ms); the vibration sensor detects the vibration amplitude during transformer operation (such as 0.5g) and transmits it through I2C.

[0070] S2: The edge intelligent node performs adaptive monitoring and dynamically adjusts the monitoring strategy The edge node runs a sliding window algorithm (window 10 seconds, step 1 second). At 14:00 one day, it detects that the transformer temperature rises to 85°C (change rate 0.6°C / second), triggers the adaptive mechanism, increases the sampling frequency of the partial discharge sensor to 2 times per second for 15 minutes, and at the same time increases the sampling of the vibration sensor to 1 time per second.

[0071] S3: The edge intelligent node performs local diagnosis, determines the device status and makes decisions The edge node loads the pre-trained SVM model (input features: temperature 85°C, partial discharge 50 pC, vibration 0.8 g), and the output result is "insulation deterioration" (probability 92%). The node immediately drives the relay (SRD-05VDC-SL-C) through GPIO to close the faulty circuit (action time 0.5 seconds), activates the audible and visual alarm (80 dB, red LED flashing), and the local flash memory records the event (timestamp 14:01:23, data size 1 KB).

[0072] S4: Upload key data to the cloud through the self-healing communication module Abnormal data (temperature 85°C, partial discharge 50 pC) is uploaded to the cloud through NB-IoT (data packet 200 bytes, time-consuming 5 seconds). At 14:05, the main channel's signal drops to -100 dBm due to nearby construction interference, and the system automatically switches to the ZigBee channel. The neighboring node (distance 50 meters) takes over the relay, and the transmission delay increases to 15 ms.

[0073] S5: The cloud platform conducts long-term data analysis and model optimization After receiving the data, the cloud stores it in InfluxDB, analyzes the temperature trend in the past 30 days (average 70°C, peak 85°C), and finds that the high temperature is related to the load peak (14:00 - 16:00). Retrain the SVM model using TensorFlow, and the accuracy rate increases to 96% after adding new data. The optimized model (800 KB) is pushed to the edge node at 4 minutes after midnight the next day (time-consuming 4 minutes).

[0074] S6: The operation and maintenance personnel monitor the equipment and issue instructions through the cloud platform The operation and maintenance personnel view the real-time status (temperature curve, partial discharge intensity) through the Web interface. After receiving the "insulation deterioration" alarm, they issue the "repair" instruction at 14:10. The instruction is transmitted to the edge node through LoRa (delay 3 seconds), and the node records the instruction and waits for the maintenance personnel to arrive for execution.

[0075] Operation effect Monitoring accuracy: The system has been running for 3 months, successfully detected 5 temperature anomalies and 2 insulation deterioration events, the false alarm rate is less than 5%, and the missed alarm rate is 0.

[0076] Response speed: The average time from anomaly detection to local decision-making is 2 seconds, which is much faster than the several-hour response of traditional manual inspections.

[0077] Communication reliability: Under the condition of simulating 50% node failures and signal interference, the data transmission success rate still reaches 92%, and the self-healing mechanism effectively avoids communication interruptions.

[0078] Energy consumption performance: The average power consumption of edge nodes is 55 mW. Solar power supply meets the continuous operation requirements, and the battery replacement cycle is extended to 3 years.

[0079] Beneficial effects

[0080] This embodiment demonstrates the efficient operation ability of the box-type substation in actual scenarios. The adaptive monitoring technology accurately captures high-temperature and insulation deterioration problems by dynamically adjusting the sampling frequency, avoiding data redundancy; the low-power edge intelligence enables local rapid diagnosis, reducing power outage losses (estimated to save 100,000 yuan / year); the self-healing communication network ensures the continuity of data transmission and adapts to the complex urban electromagnetic environment. The system significantly improves power supply reliability, reduces maintenance costs, and is suitable for wide promotion in urban distribution networks.

[0081] Table 1:

[0082] Explanation of table characters 1. System composition Embodiment of the present invention: Sensor network, edge node, self-healing communication, cloud It means that the system consists of a four-layer architecture: the sensor network collects data, the edge node performs local processing, the self-healing communication ensures data transmission, and the cloud platform is responsible for analysis and management.

[0083] Prior art 1: Traditional manual inspection system: Handheld thermometer, partial discharge detector It means relying only on handheld devices, such as thermometers and partial discharge detectors, to complete monitoring through manual operation.

[0084] Prior art 2: Fixed-parameter Internet of Things monitoring system: Fixed sensors, central server, wireless communication The system includes fixed sensors, a central server for data processing, and a simple wireless communication module.

[0085] 2. Sensor types and deployment Embodiment of the present invention: 5 types (10 pieces) The system uses 5 types of sensors (temperature, partial discharge, vibration, gas, environment), with a total of 10 deployed, covering multiple key points.

[0086] Prior art 1: Single (temporary) Only uses a single type of sensor (such as temperature or partial discharge), and is temporarily carried to the site for use.

[0087] Prior art 2: 2 - 3 types (fixed) Deploys 2 to 3 types of fixed sensors (such as temperature and partial discharge), with limited quantity and coverage.

[0088] 3. Data acquisition method Embodiment of the present invention: Adaptive (1 - 2 times per second) The data acquisition frequency is dynamically adjusted according to the device status, with a default of 1 time per second and up to 2 times per second during anomalies.

[0089] Prior art 1: Manual (1 - 2 times per day) Manual timed acquisition, usually 1 to 2 times per day, without real-time performance.

[0090] Prior art 2: Fixed (1 time per minute) The data acquisition frequency is fixed at 1 time per minute and cannot be adjusted according to changes in the situation.

[0091] 4. Data processing and diagnosis Embodiment of the present invention: Edge diagnosis (0.5 seconds) Data is processed and diagnosed locally at the edge node, and the completion time is only 0.5 seconds.

[0092] Prior art 1: Manual analysis Data needs to be manually recorded and analyzed, without an automated diagnosis function.

[0093] Prior art 2: Cloud processing (5 - 10 seconds) Data is uploaded to the cloud for processing, with a delay between 5 and 10 seconds and without local decision-making ability.

[0094] 5. Communication method Embodiment of the present invention: Self-healing (success rate 92%) Using a self-healing communication mechanism (such as channel switching and path re-planning), the data transmission success rate reaches 92%.

[0095] Prior art 1: No communication There is no communication function, and data is transmitted manually by recording.

[0096] Prior art 2: Single mode (success rate < 80%) Using a single communication mode (such as 4G), without self-healing function, and the success rate is less than 80%.

[0097] 6. Cloud functions Embodiment of the present invention: Storage, analysis, optimization The cloud provides data storage, trend analysis, and model optimization functions.

[0098] Prior art 1: None There is no cloud support, and all operations rely on manual labor.

[0099] Prior art 2: Storage, visualization The cloud only provides basic data storage and visualization functions, without advanced analysis or optimization.

[0100] 7. Response Time Embodiment of the present invention: < 20 seconds The total time from anomaly detection to decision-making and instruction issuance is less than 20 seconds.

[0101] Prior art 1: Several hours It takes several hours from inspection to problem discovery to taking measures.

[0102] Prior art 2: 10 - 30 seconds The response time is between 10 and 30 seconds, depending on the network condition.

[0103] 8. Monitoring Accuracy Embodiment of the present invention: False alarm < 5%, Miss rate 0 The false alarm rate is less than 5%, and the miss rate is 0, with highly accurate monitoring.

[0104] Prior art 1: Miss rate about 20% The miss rate is about 20%, and the accuracy depends on manual experience.

[0105] Prior art 2: False alarm 10 - 15%, Miss rate 5 - 10% The false alarm rate is 10% to 15%, and the miss rate is 5% to 10%, with medium accuracy.

[0106] 9. Power Consumption and Power Supply Embodiment of the present invention: 55 mW, Battery for 3 years The average power consumption is 55 milliwatts, and the battery life reaches 3 years, supporting solar power supply.

[0107] Prior art 1: No power consumption There is no power consumption of electronic devices, relying entirely on manual labor.

[0108] Prior art 2: 200 mW, Battery for 1 - 2 years The average power consumption is 200 milliwatts, and the battery life is 1 to 2 years, requiring external power supply support.

[0109] 10. Communication Reliability Embodiment of the present invention: 92% The data transmission success rate reaches 92% in a complex environment.

[0110] Prior art 1: None There is no communication function, not applicable to this item.

[0111] Prior art 2: < 80% The communication success rate is less than 80%, and it is vulnerable to interference.

[0112] 11. Deployment and Maintenance Cost Embodiment of the present invention: 20,000 + 5,000 / year (saving 100,000 / year) The initial deployment cost is 20,000 yuan, the annual maintenance cost is 0.5 ten thousand yuan, and the failure prevention saves the power outage loss of 1 million yuan per year.

[0113] Existing technology 1: 30,000 yuan / year The average annual labor cost is 30,000 yuan, and there is no preventive benefit.

[0114] Existing technology 2: 15,000 yuan + 10,000 yuan / year The initial cost is 15,000 yuan, the annual maintenance cost is 10,000 yuan, and there is no significant savings.

[0115] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An Internet of Things type box-type substation based on online monitoring, characterized in that, The box-type substation includes a sensor network layer, an edge intelligent node layer, a self-healing communication module layer, and a cloud management platform layer; The sensor network layer is used to collect the operating status and environmental parameters of the equipment in the box-type substation in real time; The edge intelligent node layer is used to process sensor data, perform adaptive monitoring and local diagnosis, and communicate with the cloud; The self-healing communication module layer is used to ensure the reliability and continuity of data transmission; The cloud management platform layer is used to receive the data uploaded by the edge nodes, perform long-term analysis, model training, and remote management.

2. The IoT-type box-type substation based on online monitoring according to claim 1, wherein The sensor network layer includes a variety of sensors, and the sensors are selected from one or more of temperature sensors, partial discharge sensors, vibration sensors, gas sensors, and environmental sensors.

3. The Internet of Things type box-type substation based on online monitoring according to claim 2, characterized in that The sensors are connected to the edge intelligent node layer by wired or wireless means, and some sensors support the sleep mode to reduce power consumption.

4. The Internet of Things type box-type substation based on online monitoring according to claim 1, characterized in that, The edge intelligent node layer includes an embedded processor, a storage unit, a communication interface, and a power management module; The embedded processor is used to run the adaptive monitoring module and the local diagnosis module; The storage unit is used to store firmware, local data, and machine learning models; The communication interface supports a variety of communication protocols, including LoRa, NB-IoT, ZigBee, and Ethernet; The power management module supports energy harvesting and low-power operation.

5. The Internet of Things type box-type substation based on online monitoring according to claim 4, characterized in that, The adaptive monitoring module adjusts the sensor sampling frequency according to the device status through a dynamic sampling algorithm; The local diagnosis module uses a machine learning model to analyze the data, judge the health status of the device, and trigger local control instructions.

6. The Internet of Things type box-type substation based on online monitoring according to claim 5, characterized in that, The machine learning model is trained in the cloud and updated to the edge node through OTA over-the-air download.

7. The Internet of Things type box-type substation based on online monitoring according to claim 1, characterized in that, The self-healing communication module layer includes a main communication module, an auxiliary communication module, and a network monitoring module; The main communication module is used to upload data on a low-power wide-area network; The auxiliary communication module is used for short-distance communication to form a network between nodes and data relaying; The network monitoring module is used to monitor the channel quality and node status in real time and detect communication anomalies.

8. The Internet of Things type box-type substation based on online monitoring according to claim 1, characterized in that, The self-healing communication module layer supports channel switching and path re-planning; When the signal strength of the main channel is lower than the threshold or the packet loss rate exceeds the standard, it automatically switches to the standby channel; Use neighboring nodes as relays to construct a temporary mesh network to bypass faulty nodes.

9. The Internet of Things type box-type substation based on online monitoring according to claim 1, characterized in that, The cloud management platform layer includes a data storage module, a data analysis module, a model training module, and a user interface; The data storage module is used to store historical monitoring data; The data analysis module is used to perform trend analysis and fault prediction; The model training module is used to train and optimize the machine learning model and push it to the edge node through OTA; The user interface is used for device status monitoring, alarm receiving, and control instruction issuing.

10. The IoT-based box-type substation based on online monitoring according to claim 1, characterized in that, The working process of the box-type substation includes: S1. The sensor network layer collects data and transmits it to the edge intelligent node layer; S2. The edge intelligent node layer performs adaptive monitoring and dynamically adjusts the monitoring strategy; S3. The edge intelligent node layer performs local diagnosis, judges the device status, and makes decisions; S4. Upload key data to the cloud through the self-healing communication module; S5. The cloud platform performs long-term data analysis and model optimization; S6. The operation and maintenance personnel monitor the device through the cloud platform and issue instructions.

Images