A remote intelligent patrol system for a substation

Abstract

This invention discloses a remote intelligent inspection system for substations, belonging to the field of intelligent substation inspection technology. The system adopts a layered distributed architecture, consisting of front-end inspection equipment, a back-end server, data storage devices, and an intelligent analysis platform. The front-end inspection equipment is deployed at the substation site, responsible for collecting image and video data from the equipment and transmitting the data to the back-end server via a network. The back-end server runs intelligent analysis software to process, analyze, and store the collected data, enabling real-time monitoring and intelligent diagnosis of the substation equipment status. This invention proposes a dual-closed-loop deep collaborative architecture of "task pre-simulation verification closed loop" and "model co-evolution closed loop." Through this dual-closed-loop collaboration, this invention improves identification robustness in complex environments, reduces false alarms and missed alarms, and enhances the interpretability and quantifiability of inspection decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent substation inspection technology, and in particular to a remote intelligent substation inspection system. Background Technology

[0002] Against the backdrop of the national strategy to actively promote the construction of a new power system, ensuring the safe and stable operation of substations is of paramount importance for the reliable power supply of the power system. Traditional substation inspections mainly rely on manual labor, which suffers from high labor costs, low inspection efficiency, significant susceptibility to environmental factors such as weather, and difficulty in detecting potential equipment defects.

[0003] Existing intelligent patrol systems typically include functions such as camera / robot data acquisition, remote monitoring, image recognition, and alarms. However, they still suffer from problems such as increased recognition errors under complex lighting / occlusion / severe weather conditions, lack of quantitative verification of patrol strategies before execution, and insufficient model generalization due to difficulties in sharing cross-site data. Therefore, a system is needed that can perform low-cost pre-testing and evaluation before task assignment and achieve cross-site collaborative optimization without data leaving the data domain. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a remote intelligent inspection system for substations. Through intelligent technology, it enables remote, automatic, and efficient inspection of substation equipment, timely detection of potential equipment faults, and improvement of substation operation and maintenance management and power supply reliability. This provides strong support for the construction of new power systems and is mainly applied to daily inspections, fault diagnosis, and key monitoring during special periods in various types of substations, with broad prospects for widespread application.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A remote intelligent inspection system for substations is disclosed. The system adopts a layered distributed architecture and consists of front-end inspection equipment, back-end servers, data storage devices, and an intelligent analysis platform. The system supports multiple deployment methods and can be flexibly selected according to the scale and actual situation of the substation. It can meet the simple deployment needs of small substations and adapt to the complex networking architecture of large substations, and has good scalability and compatibility. The front-end inspection equipment is deployed at the substation site and is responsible for collecting image and video data of the equipment and transmitting the data to the back-end server via the network. The back-end server runs intelligent analysis software to process, analyze and store the collected data, thereby realizing real-time monitoring and intelligent diagnosis of the substation equipment status. The data storage device is used to store the processed data and system configuration information. The intelligent analysis platform is deployed on the backend server and is used to perform intelligent analysis on the received data. The intelligent analysis platform includes a device status intelligent identification module, a remote monitoring and operation module, an automatic inspection task management module, an alarm and notification module, and a data statistics and analysis module. The intelligent equipment status recognition module is configured to use deep learning algorithms to process the collected image data in order to identify equipment instrument readings, switch opening and closing status, and appearance defects. The remote monitoring and operation module is configured to provide the system with real-time video monitoring function, and maintenance personnel can view the on-site status of substation equipment at any time through a remote terminal. The automatic inspection task management module is configured to generate and distribute inspection tasks to the front-end inspection equipment according to a preset inspection strategy, and manage the execution, suspension and modification of the tasks. The alarm and notification module is configured to trigger alarms based on the output results of the device status intelligent identification module, according to preset alarm rules, and notify maintenance personnel through various communication methods. The data statistics and analysis module is configured to perform multi-dimensional statistics and trend analysis on historical inspection data. The digital twin simulation module works in conjunction with the automatic inspection task management module and the equipment status intelligent identification module. It is configured to call the current identification model of the equipment status intelligent identification module before the physical inspection task is actually issued and executed by the automatic inspection task management module. In a digital twin scenario constructed based on the substation's 3D model, historical inspection data, and real-time environmental data, it simulates and pre-runs the execution process and identification results of the inspection task, and outputs a simulation evaluation report including the expected identification accuracy, potential missed and false alarm points, task time, and risk level. This report guides the optimization of the inspection task. The simulation pre-run of the digital twin simulation module includes: environmental disturbance simulation, strategy comparison and evaluation, and defect sample generation. The environmental disturbance simulation includes at least the injection of variables such as illumination, rain, snow, fog, haze, and occlusion. The strategy comparison and evaluation includes at least the comparison of different preset angles, different robot travel routes, and different shooting sequences. The federated learning model update module works in collaboration with the equipment status intelligent identification module and the digital twin simulation module. It is configured to, under the premise of ensuring that the original local data of each substation does not leave the domain, aggregate model parameters with other substation nodes through encryption based on the parameter updates generated by the local identification model of each station in real inspection and digital twin simulation, so as to collaboratively optimize the deep learning identification model. The digital twin simulation module is further configured to use the augmented data generated by simulation to participate in the local training process of federated learning, forming a collaborative closed loop of "simulation data generation - federated learning model update - model feedback to improve simulation evaluation accuracy". Before parameter aggregation, the federated learning model update module calculates the contribution weight of the model update uploaded by each node, and assigns higher weights to nodes that provide high-value rare defect samples or high-quality simulation augmented samples, so as to improve the global model's ability to identify rare defects.

[0006] This invention analyzes data collected by front-end devices using intelligent algorithms, accurately identifying various status information of substation equipment, such as instrument readings, switch positions, and equipment appearance defects, and enabling remote control and parameter adjustment of the equipment. Simultaneously, the system can automatically execute inspection tasks according to preset inspection strategies, promptly issuing alarms for abnormal situations and generating detailed inspection reports and data analysis reports. This provides a scientific basis for substation operation and maintenance decisions, effectively improving the efficiency and intelligence level of substation operation and maintenance.

[0007] Preferably, the intelligent equipment status identification module utilizes deep learning algorithms and image recognition technology to accurately identify and analyze key components of substation equipment such as instruments, switches, and disconnectors. The intelligent equipment status identification module can automatically read the values ​​of instruments such as SF6 pressure gauges and oil level gauges to determine whether the equipment is operating normally. It can also detect defects such as oil leakage, component damage, and corrosion by analyzing the equipment's appearance image. The identification accuracy is high, which can effectively replace manual on-site inspection and reduce the workload and errors of manual inspection.

[0008] Preferably, the remote monitoring and operation module provides the system with real-time video monitoring functionality, allowing maintenance personnel to view the on-site status of substation equipment at any time via a remote terminal, as if they were physically present. It also supports remote control of front-end inspection equipment, such as adjusting camera angles and focal lengths, and controlling robot movement. Furthermore, it enables remote operation of some equipment within the substation, such as remotely controlling switch opening and closing, and adjusting transformer tap changes. The remote monitoring and operation module achieves intelligent remote management of the substation, improving the flexibility and response speed of maintenance. It provides a graphical user interface, allowing maintenance personnel to remotely view on-site video in real time, control the movement and shooting parameters of front-end inspection equipment, and perform opening and closing control or parameter adjustment on remotely operable equipment within the substation.

[0009] Preferably, the automatic inspection task management module supports "fast inspection mode" and "deep inspection mode", and switches between the two modes based on the risk level and expected accuracy of the simulation evaluation report as trigger conditions; The automatic inspection task management module allows users to flexibly set parameters such as cycle, time, route, and inspection content for automatic inspection tasks based on the equipment distribution, operating characteristics, and maintenance requirements of the substation. The system will automatically start the inspection process according to the preset task plan. The front-end inspection equipment will automatically collect data according to the specified route and inspection points. The back-end system will synchronously process and analyze the data and automatically record and store the inspection results. At the same time, the system supports operations such as pausing, continuing, modifying, and terminating tasks, and can adjust the inspection strategy in a timely manner according to the actual situation to ensure the efficiency and adaptability of the inspection work.

[0010] Preferably, the alarm and notification module is mainly used when the system detects an abnormality or fault in the equipment. It will immediately trigger the alarm mechanism to generate alarm information and notify the operation and maintenance personnel in a timely manner through various means such as sound, pop-up window, SMS, and email. The alarm information includes key information such as the detailed location of the abnormal equipment, the type of abnormality, and the severity, which helps the operation and maintenance personnel to quickly locate the fault point, shorten the fault response time, improve the fault handling efficiency, and reduce the impact of equipment failure on the power grid operation.

[0011] Preferably, the data statistics and analysis module enables the system to have powerful data statistics and analysis functions, and can perform multi-dimensional statistical analysis on historical inspection data, such as classifying and statistically analyzing by equipment type, inspection cycle, abnormal situation, etc., and generating various reports and charts. Through in-depth data mining, the operating trend and fault pattern of the equipment can be analyzed, providing data support for equipment maintenance, upgrading and transformation, and optimization of operation and maintenance strategies, so as to realize the scientific and refined operation and maintenance management of substations.

[0012] Preferably, the intelligent device status recognition module further includes: The preset position configuration submodule is used to associate camera preset positions with specific inspection points. The preset positions include pre-set camera angle and focal length parameters. The identification configuration submodule is used to configure the identification type and calculation method for a specific device area under each preset position. The identification type includes device appearance inspection, meter identification, and line disconnector identification. The calculation method includes conventional remote signaling and upper and lower limit calculation.

[0013] Preferably, the alarm and notification module supports multiple alarm calculation logics, including: Upper and lower limit alarms: triggered when the detected value exceeds the preset upper or lower limit; Expression alert: Calculated and triggered based on a user-defined logical expression; Appearance-related alarms: triggered when the identification type is device appearance inspection and a defect is identified; The alarm and notification module is also configured to: if the identification result is marked as failure or the location is set to the whitelist, then skip the alarm calculation, that is, execute the whitelist filtering and identification failure degradation strategy before alarm calculation.

[0014] Preferably, the front-end inspection equipment includes an inspection robot, a fixed high-definition PTZ camera, and environmental sensors. The back-end server communicates with the front-end inspection equipment through a combination of wired and wireless networks and is equipped with a network security protection mechanism.

[0015] The beneficial effects of this invention are as follows: This invention proposes a dual-loop deep collaborative architecture of "task pre-simulation verification closed loop" and "model co-evolution closed loop": (1) Task pre-simulation verification closed loop: Before issuing physical tasks, the automatic inspection task management module calls the digital twin simulation module to perform a pre-run, outputs a quantitative evaluation report, and automatically adjusts the route / preset position / timing based on the report to achieve strategy optimization from "blind ordering" to "precise calculation"; (2) Model co-evolution closed loop: The federated learning model update module integrates real inspection samples and simulation augmented samples for local training, and encrypts and aggregates them to form a global model; after the global model is issued, it improves the recognition accuracy and feeds back the simulation evaluation accuracy, forming a spiral upward self-optimization mechanism.

[0016] Through the above-mentioned dual closed-loop collaboration, the present invention improves the robustness of identification in complex environments, reduces false alarms and missed alarms, and enhances the interpretability and quantifiability of inspection decisions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the composition of a remote intelligent inspection system for substations according to the present invention; Figure 2 This is a schematic diagram of the composition of the intelligent analysis platform of the present invention; Figure 3 This is a schematic diagram of the preset position configuration interface of the system of the present invention; Figure 4 This is a schematic diagram of the interface for configuring the appearance recognition type of the device of the present invention; Figure 5 This is a schematic diagram of the meter identification configuration interface of the present invention; Figure 6 This is an example of the conventional instrument (line switch) identification configuration method of the present invention. Figure 1 ; Figure 7 This is an example of the conventional instrument (line switch) identification configuration method of the present invention. Figure 2 ; Figure 8This is an example of the conventional instrument (line switch) identification configuration method of the present invention. Figure 3 ; Figure 9 This is an example diagram of the rotary switch configuration method of the present invention; Figure 10 This is a schematic diagram of the interface for creating a new patrol task in this invention; Figure 11 This is a schematic diagram of the newly created inspection host interface of the present invention; Figure 12 This is a schematic diagram of the inspection results viewing interface of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1-2 A remote intelligent inspection system for substations is proposed. The system adopts a layered distributed architecture and consists of front-end inspection equipment, back-end servers, data storage devices, and an intelligent analysis platform. The system supports multiple deployment methods and can be flexibly selected according to the scale and actual situation of the substation. It can meet the simple deployment needs of small substations and adapt to the complex networking architecture of large substations, and has good scalability and compatibility.

[0020] Specifically, the front-end inspection equipment is deployed at the substation site to collect images, videos, and other data from the equipment. This data is then transmitted to the back-end server via a network. The back-end server runs intelligent analysis software to process, analyze, and store the collected data, enabling real-time monitoring and intelligent diagnosis of the substation equipment status. Data storage devices store the processed data and system configuration information. The intelligent analysis platform, deployed on the back-end server, performs intelligent analysis on the received data. As a further optimization, the front-end inspection equipment includes an inspection robot, a fixed high-definition PTZ camera, and environmental sensors. The back-end server communicates with the front-end inspection equipment via a combination of wired and wireless networks, and a network security protection mechanism is in place.

[0021] Specifically, the intelligent analysis platform includes a device status intelligent identification module, a remote monitoring and operation module, an automatic inspection task management module, an alarm and notification module, and a data statistics and analysis module. The intelligent equipment status identification module is configured to process the collected image data using deep learning algorithms to identify equipment instrument readings, switch opening and closing status, and appearance defects. As a further optimization solution, the intelligent equipment status identification module uses deep learning algorithms and image recognition technology to accurately identify and analyze key parts of substation equipment such as instruments, switches, and disconnectors. The intelligent equipment status identification module can automatically read the values ​​of instruments such as SF6 pressure gauges and oil level gauges to determine whether the equipment is operating normally. It can also detect defects such as oil leakage, component damage, and corrosion by analyzing the equipment appearance images. The identification accuracy is high, which can effectively replace manual on-site inspection and reduce the workload and error of manual inspection. The remote monitoring and operation module is configured to provide the system with real-time video monitoring capabilities, allowing maintenance personnel to view the on-site status of substation equipment at any time via remote terminals. As a further optimization, the remote monitoring and operation module provides the system with real-time video monitoring capabilities, allowing maintenance personnel to view the on-site status of substation equipment at any time via remote terminals, as if they were there in person. It supports remote control of front-end inspection equipment, such as adjusting camera angles and focal lengths, controlling robot movement, etc. It can also realize remote operation of some equipment within the substation, such as remotely controlling the opening and closing of switches, adjusting transformer taps, etc. The remote monitoring and operation module realizes intelligent remote management of the substation, improves the flexibility and response speed of operation and maintenance, provides a graphical operation interface, and allows maintenance personnel to remotely view on-site video in real time, control the movement and shooting parameters of front-end inspection equipment, and perform opening and closing control or parameter adjustment of equipment within the substation that supports remote operation. The automatic inspection task management module is configured to generate and distribute inspection tasks to front-end inspection equipment according to preset inspection strategies, and manage the execution, suspension, and modification of tasks. As a further optimization solution, the automatic inspection task management module allows users to flexibly set parameters such as the cycle, time, route, and inspection content of automatic inspection tasks according to the equipment distribution, operating characteristics, and maintenance requirements of the substation. The system will automatically start the inspection process according to the preset task plan. The front-end inspection equipment will automatically collect data according to the specified route and inspection points. The back-end system will synchronously process and analyze the data, and automatically record and store the inspection results. At the same time, the system supports operations such as pausing, continuing, modifying, and terminating tasks, and can adjust the inspection strategy in a timely manner according to the actual situation to ensure the efficiency and adaptability of the inspection work. The alarm and notification module is configured to trigger alarms based on the output of the intelligent device status recognition module and preset alarm rules, and notify maintenance personnel through various communication methods. As a further optimization, the alarm and notification module mainly triggers the alarm mechanism to generate alarm information immediately when the system detects device abnormalities or faults. It promptly notifies maintenance personnel through various means such as sound, pop-up window, SMS, and email. The alarm information includes key information such as the detailed location of the abnormal device, the type of abnormality, and the severity, helping maintenance personnel to quickly locate the fault point, shorten the fault response time, improve fault handling efficiency, and reduce the impact of equipment failure on power grid operation. The data statistics and analysis module is configured to perform multi-dimensional statistics and trend analysis on historical inspection data. As a further optimization solution, the data statistics and analysis module enables the system to have powerful data statistics and analysis functions, and can perform multi-dimensional statistical analysis on historical inspection data, such as classifying and statistically analyzing by equipment type, inspection cycle, abnormal situation, etc., generating various reports and charts. Through in-depth data mining, the operating trends and fault patterns of equipment can be analyzed, providing data support for equipment maintenance, upgrading and transformation, and optimization of operation and maintenance strategies, thereby realizing the scientific and refined operation and maintenance management of substations.

[0022] This embodiment analyzes data collected by front-end devices using intelligent algorithms, accurately identifying various status information of substation equipment, such as instrument readings, switch positions, and equipment appearance defects, and enabling remote control and parameter adjustment of the equipment. Simultaneously, the system can automatically execute inspection tasks according to preset inspection strategies, promptly issuing alarms for abnormal situations and generating detailed inspection reports and data analysis reports. This provides a scientific basis for substation operation and maintenance decisions, effectively improving the efficiency and intelligence level of substation operation and maintenance.

[0023] Example 2 is an optimization based on Example 1. Specifically, the intelligent device status recognition module further includes: The preset position configuration submodule is used to associate camera preset positions with specific inspection points. The preset positions include pre-set camera angle and focal length parameters. The identification configuration submodule is used to configure the identification type and calculation method for a specific equipment area under each preset position. The identification types include equipment appearance inspection, meter identification, and line disconnector identification. The calculation methods include conventional remote signaling and upper and lower limit calculation.

[0024] In the specific implementation process, refer to Figure 3As shown, the page location is: Configuration Management - Preset Position Configuration: In area 1, select the cameras corresponding to the inspection points; in area 7, select the points in the inspection point table that need to have preset positions created; in area 2, select the corresponding camera (pre-set the shooting angle in the streaming media and point it at the on-site equipment to be filmed); in area 5, select the preset position number as the preset position type, and enter the unused preset position number in the camera; in area 6, click Settings, and then click Save; if you need to configure the configuration or use the preset position correction function, select the preset position point in area 4, and then click Screenshot.

[0025] In the specific implementation process, refer to Figure 4 As shown, the appearance defect recognition settings - meter recognition configuration requires configuration for all non-device appearance inspections. First, click the recognition configuration in the upper right corner, then double-click the preset position created in ① to load the screenshot file of the preset position configuration, and then click ② to select the corresponding point.

[0026] In the specific implementation process, refer to Figure 5 As shown, use the black box that appears to frame the device to be identified in the middle image 3. Then, click to draw the calculation area in the small frame on the right (4), or press the "space" key to zoom in (e.g., for the meter shown in the image, first frame the center point in the center of the pointer (5), then frame each large scale with a small frame; in the image, the meter starts at 0 and goes up to 5). After framing, first select the identification type in 6, then select the calculation method in 7, then click "Save" above the small image (8), and finally click "Save" again in the top (9).

[0027] In the specific implementation process, refer to Figure 6 As shown, the standard instrument identification is as follows: Identification Type - Line Switch / Line. Calculation Method - Standard Remote Signaling (Default Closed) / Standard Remote Signaling (Default Disconnected). Configuration Method: When the line switch has one fixed end and one retractable end, select the fixed end during configuration; when the line switch can rotate on both sides, select the connection part if it is closed, and select one side if it is open.

[0028] In addition, when the disconnect switch is in the open position, there are two configuration methods, as follows: One is as follows Figure 7 As shown, select the approximate closing position in the configuration box, and select the group "Conventional Remote Signaling (Default)" for the calculation method; secondly, as... Figure 8 In the configuration box, select the tripping position and choose "Conventional Remote Signaling (Default Closed)" as the calculation method.

[0029] The configuration method for rotary switch is as follows: Figure 9 As shown, the connected parts are selected; the identification type is line disconnect switch / straight line; the calculation method is conventional remote signaling.

[0030] Example 3 is an optimization based on Example 1. Specifically, the alarm and notification module supports multiple alarm calculation logics, including: Upper and lower limit alarms: triggered when the detected value exceeds the preset upper or lower limit; Expression alert: Calculated and triggered based on a user-defined logical expression; Appearance-related alarms: triggered when the identification type is device appearance inspection and a defect is identified; The alarm and notification module is also configured to skip alarm calculation if the identification result is marked as failed or the location is set to a whitelist. The system has added a dynamic alarm threshold adjustment algorithm that integrates environmental data and historical equipment operating data. The dynamic threshold adjustment submodule within the alarm module can access real-time data from environmental sensors (temperature, humidity, pollution level) and the SCADA system, including historical equipment load and operating temperature data. Through a pre-trained correlation model, this submodule can dynamically calculate and adjust alarm thresholds under different operating conditions (e.g., appropriately increasing the upper limit of transformer oil temperature alarms under high temperature and high load conditions), making alarms more intelligent and accurate, and reducing false alarms and missed alarms.

[0031] Example 4 is an optimization based on Example 1. Specifically, after in-depth practice, the inventors of this application discovered that building a truly efficient and reliable remote intelligent inspection system for substations faces two deeper, interconnected "decision-making closed-loop" bottlenecks: 1. Decision-making risks caused by the disconnect between planning and verification: Existing automated inspection task management relies on preset fixed strategies (routes, pre-defined locations). Before actual execution, maintenance personnel cannot predict the feasibility and accuracy of the task in complex field environments (such as changes in lighting, equipment obstruction), nor can they assess the potential benefits and costs of different inspection strategies (such as adjusting routes, adding pre-defined locations). This leads to blind inspection planning, potentially resulting in ineffective inspections or missed inspections. Essentially, it lacks a low-cost, zero-risk, and quantifiable "pre-implementation sandbox" for physical inspection tasks in a virtual space.

[0032] 2. Evolutionary Stagnation Due to "Data-Model" Isolation: Individual substations have few defect samples, and each substation forms a "data silo" due to differences in equipment models and operating environments. Identification models trained on single-station data have weak generalization capabilities and struggle to cope with diverse equipment states and emerging defect types. Furthermore, due to the high sensitivity of power data, it is impossible to centrally train a universal model using data from various substations. This leads to a negative cycle in the system's core identification capabilities: "few data - poor model - inaccurate identification - even less effective data," lacking a feasible path to achieve cross-station knowledge sharing and model co-evolution while ensuring data sovereignty and security.

[0033] Therefore, a remote intelligent inspection system for substations also includes a digital twin simulation module. The digital twin simulation module works in conjunction with the automatic inspection task management module and the equipment status intelligent identification module. It is configured to call the current identification model of the equipment status intelligent identification module before the physical inspection task is actually issued and executed by the automatic inspection task management module. In the digital twin scenario constructed based on the substation 3D model, historical inspection data and real-time environmental data, the execution process and identification results of the inspection task are simulated and pre-performed. The system outputs a simulation evaluation report including the expected identification accuracy, potential missed and false alarm points and task time, which is used to guide the optimization of the inspection task. The system integrates a digital twin simulation module, which is based on a three-dimensional digital twin model of the substation. Before the actual execution of the physical inspection task, it can simulate and rehearse the preset inspection route and identification actions. It can combine historical identification data and equipment status to predict possible identification results and potential alarm points, helping maintenance personnel to evaluate the rationality and effectiveness of the inspection plan in advance and optimize task configuration. Specifically, the simulation preview of the digital twin simulation module includes: Environmental disturbance simulation: In the digital twin scenario, simulated variables such as illumination at different times, rain, snow, fog and haze, and seasonal vegetation occlusion are injected to evaluate their impact on image acquisition quality and subsequent recognition accuracy. Strategy comparison and evaluation: For the same inspection target, simulate various task strategies with different preset position angles, different inspection robot travel routes, and different shooting sequences, and compare the simulation evaluation reports under each strategy to select the optimal strategy; Defect sample generation: Based on the 3D model of the equipment and the defect generation algorithm, sample images of various rare defects or extreme conditions are automatically synthesized in the simulation environment as augmented data output.

[0034] In this embodiment, the "environmental disturbance simulation" and "strategy comparison and evaluation" directly solve the "plan-verification disconnect" problem; while the "defect sample generation" provides a key data supplementation method to solve the "data-model isolation" problem.

[0035] A remote intelligent inspection system for substations also includes a federated learning model update module, which is connected to an intelligent equipment status recognition module. The federated learning model update module works in collaboration with the equipment status intelligent identification module and the digital twin simulation module. It is configured to, under the premise of ensuring that the original local data of each substation does not leave the domain, aggregate model parameters with other substation nodes through encryption based on the parameter updates generated by the local identification model of each station in real inspection and digital twin simulation, so as to collaboratively optimize the deep learning identification model. Furthermore, the digital twin simulation module is configured to use the augmented data generated by simulation to participate in the local training process of federated learning.

[0036] The workflow of the federated learning model update module includes: Local training data source fusion: Local training data for each substation node, including samples collected from real inspections and labeled simulation augmented samples generated by the digital twin simulation module; Differentiated contribution assessment: Before parameter aggregation, the contribution weight of the model update uploaded by each node is calculated. Among them, nodes that provide high-value simulation defect samples or real rare defect samples are given higher weight. Model effectiveness feedback: After the aggregated global model is distributed to each node, its performance improvement will be fed back to the digital twin simulation module to adjust the focus of its defect generation algorithm.

[0037] The system introduces a multi-station collaborative adaptive identification model update mechanism based on federated learning. The system includes a federated learning model update module, which allows instances of the system deployed in different substations to aggregate and update model parameters (such as gradients and weights) through encrypted transmission without exchanging original local data. This enables the identification model to continuously optimize using a wider range of data samples, improving its generalization ability and identification accuracy across various devices and environmental conditions, while strictly adhering to data security and privacy protection requirements.

[0038] Because this embodiment innovatively constructs a dual-closed-loop deep collaborative architecture of "digital twin simulation verification" and "federated learning co-evolution," it provides scarce data for federated learning through simulation, and the improved model from federated learning in turn feeds back into the simulation accuracy, thereby generating a mutually reinforcing and spiraling synergistic effect, achieving the following unexpected technical results: 1. It has enabled the leap from "blind ordering" to "precise calculation" of inspection tasks: Because the digital twin simulation module can perform multi-strategy and multi-environment pre-playing and quantitative evaluation of tasks in a virtual environment, maintenance personnel can select the most efficient and reliable inspection plan under the premise of zero risk and low cost, which reduces the risk of invalid inspection and missed inspection from the root. This is something that simple task management or independent digital twin display cannot achieve.

[0039] 2. Overcoming the model evolution dilemma under the limitation of single substation data: Because the federated learning model update module innovatively integrates real data and simulation-generated augmented data, and introduces a contribution evaluation mechanism, the system can effectively utilize cross-station knowledge under the premise of strictly ensuring data security. In particular, it can achieve continuous and collaborative evolution of the identification model for rare defects that are scarce in each station but have great harm, which significantly improves the overall robustness and adaptability of the system.

[0040] 3. The system has formed an endogenous driving force for self-optimization: The two closed loops mentioned above are coupled with each other. The improvement of simulation accuracy depends on a more accurate recognition model, while a more accurate model requires higher quality simulation data for training. This synergistic driving relationship enables the system to have an endogenous driving force for continuous self-improvement, and the level of intelligence can be continuously improved over time.

[0041] Example 5: This example discloses that the inspection task of this system includes the following steps: To create a new inspection task in S1: Click the Intelligent Inspection module, select Inspection Plan Management, and you will enter the following interface. Select Create Inspection Task, as shown below. Figure 10 As shown, select the corresponding plant, inspect the host, and select the inspection points in this inspection task. Fill in the basic information on the right according to the actual needs on site. After completion, click submit, and then click execute now to carry out the inspection (if it is a regional access edge node, a task plan needs to be issued. You must enter the username and password before you can click execute now). Copy the inspection host folder that comes with the server to the same directory. S2 newly built inspection host, such as Figure 11 Modify directory configuration file - modify startup script - grant file permissions - start patrol host service; S3 View the inspection results as follows Figure 12 As shown: After execution, you can click to view the inspection results in the inspection task management section under intelligent inspection to check if the results are correct.

[0042] The operation and maintenance requirements for this system include the following: Hardware requirements: The front-end inspection equipment must have high-resolution image acquisition capabilities, a stable communication module, and sufficient storage capacity to ensure accurate data acquisition and transmission. The back-end server needs to be configured with appropriate CPU, memory, disk, and other hardware resources based on the system's scale and data volume to ensure stable system operation and efficient data processing. Additionally, reliable data storage devices are required to meet the needs of long-term data preservation and backup.

[0043] Network Environment Requirements: The system has high requirements for network bandwidth and stability to ensure smooth and latency-free data transmission between front-end devices and back-end servers. Wired network connections are recommended. For remote areas or special scenarios, wireless networks can be used as a supplement, but the signal strength and transmission quality of the wireless network must meet the system requirements. In addition, a network security protection mechanism must be established to prevent external attacks and data leaks, ensuring the secure operation of the system.

[0044] Software environment requirements: The backend server must have an operating system, database management system, intelligent analysis software, and other necessary software installed, ensuring software compatibility and stability. Additionally, it must be equipped with appropriate system management tools and security software for daily system maintenance and security management.

[0045] Personnel Requirements: Maintenance personnel must possess certain professional knowledge of electrical engineering, computer operation skills, and system maintenance experience. They must be proficient in the system's operation methods and functions, handle system alarm information promptly and accurately, and perform basic system maintenance and troubleshooting. Furthermore, professional technicians must be assigned to conduct regular inspections and maintenance of the system to ensure its long-term stable operation.

[0046] The key technologies researched in this system include the following: 1. Intelligent Recognition Algorithm Research: In-depth research on deep learning-based image recognition, target detection, and semantic segmentation algorithms. Optimizing algorithm models based on the characteristics and operating environment of substation equipment to improve the accuracy and robustness of equipment status feature recognition. Simultaneously, research on multi-source data fusion technology will be conducted, combining image data with equipment operation data and environmental data to achieve a more comprehensive and accurate assessment of equipment status. Furthermore, the key technical indicator of intelligent recognition accuracy is: an equipment status intelligent recognition accuracy rate of no less than 95%, with an instrument reading recognition accuracy rate of over 98% and an equipment appearance defect recognition accuracy rate of over 90%, meeting the accuracy requirements for equipment status monitoring in daily substation operation and maintenance. 2. Remote Monitoring and Operation Technology Development: Development of highly stable remote communication protocols and control command transmission mechanisms to ensure real-time and reliable data interaction and control command execution between front-end inspection equipment and back-end servers. A user-friendly remote monitoring operation interface will be designed to enable intuitive and convenient control of front-end equipment, as well as remote parameter configuration and operation functions for substation equipment, improving the system's remote control performance and user experience. In addition, key technical indicators include system stability and reliability: the system possesses high stability, with an annual mean time between failures (MTBF) of no less than 10,000 hours, enabling stable operation in complex substation environments and ensuring continuous monitoring and management of substation equipment. Simultaneously, the system has a robust fault-tolerance mechanism and data backup and recovery capabilities, guaranteeing data security and integrity. 3. Automatic Inspection Task Management System Design: A flexible and configurable automatic inspection task management module is constructed, and task scheduling algorithms are developed to achieve intelligent management and optimized scheduling of inspection tasks. Inspection route planning algorithms are researched to generate optimal inspection routes based on substation layout and equipment distribution, improving inspection efficiency and reducing equipment energy consumption. Furthermore, key technical indicators include inspection efficiency: compared to traditional manual inspection methods, the system's automatic inspection efficiency is increased by more than 3 times. In substations of the same scale, inspection time is reduced to 1 / 3 of the original, significantly improving substation operation and maintenance efficiency and reducing the workload and time costs of manual inspection. 4. Alarm and Notification Mechanism Optimization: Establish a comprehensive alarm rule engine, formulate reasonable alarm thresholds and alarm levels based on equipment status characteristics and operating parameters, and achieve accurate alarms for equipment anomalies. Develop multiple alarm notification channels and methods, and ensure timely and accurate delivery of alarm information, enabling maintenance personnel to obtain equipment fault information and take corresponding measures immediately. In addition, the key technical indicator alarm response time: the time from detecting equipment anomaly to issuing an alarm notification should not exceed 5 seconds, ensuring that maintenance personnel can obtain fault information in a timely manner, respond quickly and handle equipment faults, and reduce the impact of faults on power grid operation. 5. Data Statistics and Analysis Model Construction: Design an efficient data storage structure and database management system to achieve efficient storage and management of massive amounts of inspection data.A data statistics and analysis model is constructed, and data mining and machine learning technologies are used to conduct in-depth analysis and mining of data, extracting valuable information and knowledge to provide strong support for substation operation and maintenance decisions. In addition, key technical indicators include data storage and processing capabilities: the system can support the storage and efficient processing of massive amounts of inspection data, with data storage capacity expandable to the PB level and data processing speed exceeding one million records per second, meeting the long-term inspection data storage and analysis needs of large substations and providing strong support for data statistics and analysis.

[0047] In summary: 1. This invention features intelligent recognition technology based on multi-source data fusion: by fusing and analyzing image data with various data sources such as equipment operation data and environmental data, it overcomes the limitations of single-source data recognition, achieving a more comprehensive and accurate assessment of substation equipment status and improving the accuracy and reliability of equipment fault diagnosis; 2. This invention develops an adaptive inspection task management system based on deep learning, which can automatically adjust the cycle, route, and inspection content of inspection tasks according to the operating status of substation equipment and historical inspection data, realizing intelligent inspection, improving the pertinence and efficiency of inspection work, and reducing operation and maintenance costs; 3. This invention adopts advanced deep learning algorithms and image recognition technology, enabling the system to reach an industry-leading level in intelligent equipment status recognition, with high recognition accuracy, and can quickly and accurately detect potential equipment faults, providing a basis for preventive maintenance of substations. It provides strong support and has significant technical advantages compared to traditional inspection methods; 4. The system of this invention has efficient remote monitoring and operation functions. Utilizing a combination of wired and wireless network communication methods, it realizes real-time, remote, and precise control of substation equipment, breaking the bottleneck of traditional substation operation and maintenance being limited by geographical location, improving the flexibility and response speed of operation and maintenance, and representing the development direction of intelligent operation and maintenance of substations; 5. Through a federated learning mechanism, this invention achieves multi-site collaborative optimization while ensuring data privacy, continuously improving the accuracy and robustness of the system's core identification model. By dynamically adjusting alarm thresholds, the system can adapt to complex and ever-changing on-site operating environments, significantly improving the accuracy and practicality of alarms. Furthermore, through digital twin simulation, it realizes the pre-verification and optimization of inspection tasks, reducing the uncertainty of on-site execution and improving the scientific nature of operation and maintenance decisions.

[0048] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A remote intelligent inspection system for substations, characterized in that, The system adopts a layered distributed architecture, consisting of front-end inspection equipment, back-end servers, data storage devices, and an intelligent analysis platform. The system supports multiple deployment methods and can be flexibly selected according to the scale and actual situation of the substation. It can meet the simple deployment needs of small substations and adapt to the complex networking architecture of large substations, and has good scalability and compatibility. The front-end inspection equipment is deployed at the substation site and is responsible for collecting image and video data of the equipment and transmitting the data to the back-end server via the network. The back-end server runs intelligent analysis software to process, analyze and store the collected data, thereby realizing real-time monitoring and intelligent diagnosis of the substation equipment status. The data storage device is used to store the processed data and system configuration information. The intelligent analysis platform is deployed on the backend server and is used to perform intelligent analysis on the received data. The intelligent analysis platform includes a device status intelligent identification module, a remote monitoring and operation module, an automatic inspection task management module, an alarm and notification module, and a data statistics and analysis module. The intelligent equipment status recognition module is configured to use deep learning algorithms to process the collected image data in order to identify equipment instrument readings, switch opening and closing status, and appearance defects. The remote monitoring and operation module is configured to provide the system with real-time video monitoring function, and maintenance personnel can view the on-site status of substation equipment at any time through a remote terminal. The automatic inspection task management module is configured to generate and distribute inspection tasks to the front-end inspection equipment according to a preset inspection strategy, and manage the execution, suspension and modification of the tasks. The alarm and notification module is configured to trigger alarms based on the output results of the device status intelligent identification module, according to preset alarm rules, and notify maintenance personnel through various communication methods. The data statistics and analysis module is configured to perform multi-dimensional statistics and trend analysis on historical inspection data. The digital twin simulation module works in collaboration with the automatic inspection task management module and the equipment status intelligent identification module. It is configured to call the current identification model of the equipment status intelligent identification module before the physical inspection task is actually issued and executed by the automatic inspection task management module. In the digital twin scenario constructed based on the substation 3D model, historical inspection data and real-time environmental data, the execution process and identification results of the inspection task are simulated and pre-run. The simulation evaluation report, which includes the expected identification accuracy, potential missed and false alarm points and task time and risk level, is output to guide the optimization of the inspection task. The federated learning model update module works in collaboration with the equipment status intelligent identification module and the digital twin simulation module. It is configured to, under the premise of ensuring that the original local data of each substation does not leave the domain, aggregate model parameters with other substation nodes through encryption based on the parameter updates generated by the local identification model of each station in real inspection and digital twin simulation, so as to collaboratively optimize the deep learning identification model. Furthermore, the digital twin simulation module is further configured to use the augmented data generated by simulation to participate in the local training process of federated learning, forming a collaborative closed loop of "simulation data generation - federated learning model update - model feedback to improve simulation evaluation accuracy".

2. The substation remote intelligent inspection system according to claim 1, characterized in that, The intelligent equipment status recognition module uses deep learning algorithms to process the collected image data to identify equipment instrument readings, switch opening and closing status, and appearance defects. It also performs multi-source consistency verification between the recognition results and equipment operation data and environmental data. When the recognition conclusion conflicts with the operating measurement point / environment threshold constraints, the alarm confidence level is reduced or it is marked as needing to be reviewed.

3. The substation remote intelligent inspection system according to claim 1, characterized in that, The remote monitoring and operation module provides the system with real-time video monitoring capabilities. Maintenance personnel can view the on-site status of substation equipment at any time through remote terminals, as if they were there in person. It also supports remote control of front-end inspection equipment and remote operation of some equipment within the substation, enabling intelligent remote management of the substation and improving the flexibility and response speed of operation and maintenance.

4. The substation remote intelligent inspection system according to claim 1, characterized in that, The automatic inspection task management module supports "fast inspection mode" and "deep inspection mode," and switches between the two modes based on the risk level and expected accuracy of the simulation evaluation report. The federated learning model update module calculates the contribution weight of the model updates uploaded by each node before parameter aggregation, and assigns higher weights to nodes that provide high-value rare defect samples or high-quality simulation augmentation samples to improve the global model's ability to identify rare defects. The simulation pre-run of the digital twin simulation module includes: environmental disturbance simulation, strategy comparison and evaluation, and defect sample generation. Among them, environmental disturbance simulation includes at least the injection of variables such as illumination, rain, snow, fog, haze, and occlusion; strategy comparison and evaluation includes at least the comparison of different preset position angles, different robot travel routes, and different shooting sequences.

5. A remote intelligent inspection system for substations according to claim 1, characterized in that, The alarm and notification module is mainly responsible for triggering an alarm mechanism to generate alarm information when the system detects equipment abnormalities or malfunctions. This information is then promptly communicated to maintenance personnel through various means, including sound, pop-ups, SMS, and email. The alarm information includes the detailed location of the abnormal equipment, the type of abnormality, and the severity of the malfunction, helping maintenance personnel to quickly locate the fault point, shorten the fault response time, improve fault handling efficiency, and reduce the impact of equipment failures on power grid operation.

6. The substation remote intelligent inspection system according to claim 1, characterized in that, The data statistics and analysis module enables the system to have powerful data statistics and analysis functions, and can perform multi-dimensional statistical analysis on historical inspection data. Through in-depth data mining, it can analyze the operating trends and fault patterns of equipment, and provide data support for equipment maintenance, upgrading and transformation, and optimization of operation and maintenance strategies, so as to realize the scientific and refined operation and maintenance management of substations.

7. A remote intelligent inspection system for substations according to claim 1, characterized in that, The intelligent device status recognition module further includes: The preset position configuration submodule is used to associate camera preset positions with specific inspection points. The preset positions include pre-set camera angle and focal length parameters. The identification configuration submodule is used to configure the identification type and calculation method for a specific device area under each preset position. The identification type includes device appearance inspection, meter identification, and line disconnector identification. The calculation method includes conventional remote signaling and upper and lower limit calculation.

8. A remote intelligent inspection system for substations according to claim 1, characterized in that, The alarm and notification module supports upper and lower limit alarms, expression alarms and appearance-based alarms, and performs whitelist filtering and identification failure degradation strategies before alarm calculation.

9. A remote intelligent inspection system for substations according to claim 1, characterized in that, The front-end inspection equipment includes an inspection robot, a fixed high-definition PTZ camera, and environmental sensors. The back-end server communicates with the front-end inspection equipment through a combination of wired and wireless networks and is equipped with a network security protection mechanism.

Images