Substation intelligent inspection auxiliary control scheduling system and method based on cloud-edge collaboration

Abstract

This invention discloses a cloud-edge collaborative intelligent substation inspection and auxiliary control scheduling system and method, belonging to the field of power system automation technology, aiming to solve the problems of imbalanced technical architecture, uneven distribution of computing power, and disconnection of collaborative logic in existing technologies. The system includes a perception and execution layer, an edge processing layer, a network transmission layer, and a cloud decision-making layer. The method includes: real-time acquisition and synchronization of substation operation data; feature extraction and instruction preprocessing using edge nodes; dynamic allocation of cloud-edge computing tasks based on deep reinforcement learning; prediction of operating trends and construction of a digital twin model through neural networks; calculation of state deviations to generate decision suggestions; and issuance of instructions to achieve closed-loop linkage between inspection and auxiliary control. Through the above solutions, this invention effectively reduces network bandwidth pressure and response latency, achieves optimal cross-domain resource allocation and predictive maintenance, and significantly enhances the operational robustness and automation level of substations in complex environments.

Description

Technical Field

[0001] This invention belongs to the field of power system automation, specifically, it relates to a cloud-edge collaborative intelligent substation inspection auxiliary control and dispatching system and method. Background Technology

[0002] As a critical infrastructure for modern society, the safe and stable operation of the power system is of paramount importance to ensuring social production and daily life. Substations, as the core hubs of the power system, bear the important functions of voltage transformation, power reception, and distribution. The operating status of their primary and secondary equipment directly affects the power supply reliability and operational safety of the entire power grid.

[0003] Intelligent inspection and auxiliary control scheduling of substations are core components for improving the automation level of substation operation and maintenance and realizing the digital transformation of the power grid. By integrating high-precision sensors, high-definition cameras and various monitoring terminals, the system can acquire environmental parameters and equipment operating status within the substation in real time, and optimize the allocation of computing resources by combining cloud-edge collaborative architecture, aiming to build an intelligent operation and maintenance system that is sensitive to perception, responds quickly, and schedules accurately.

[0004] Traditional substation inspection technologies suffer from several drawbacks: First, the data processing architecture is unbalanced, relying excessively on the cloud center for full data processing. This leads to significant network bandwidth pressure during concurrent inspections of multiple devices and results in high response latency due to long transmission links, making it difficult to meet the real-time auxiliary control requirements under emergency substation conditions. Second, uneven distribution of computing power at the edge, coupled with a lack of dynamic segmentation and scheduling mechanisms for complex inspection tasks, makes edge nodes prone to performance bottlenecks when processing high-density image recognition or multi-dimensional feature fusion. Third, the collaborative logic between the cloud and the edge lacks deep coupling, and the feedback mechanisms for data synchronization and command issuance are inadequate, resulting in a logical disconnect between inspection tasks and auxiliary control actions, making it difficult to achieve closed-loop linkage response to substation anomalies. Finally, the system lacks robustness to complex electromagnetic environments and variable communication quality, and lacks adaptive computation offloading and task migration strategies. These issues collectively contribute to low substation inspection efficiency and deviations in scheduling command execution, becoming bottlenecks hindering the development of unmanned substation operation and maintenance. Summary of the Invention

[0005] The purpose of this invention is to provide a cloud-edge collaborative intelligent inspection and control scheduling system and method for substations, which mainly solves the problems of imbalance in existing technical architecture, uneven distribution of computing power, and disconnection of collaborative logic.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A cloud-edge collaborative intelligent substation inspection auxiliary control and dispatch system includes:

[0008] The perception and execution layer is used to construct a multi-dimensional perception network and collect substation operation data in real time. The perception and execution layer is deployed at the substation site and includes industrial-grade high-definition cameras, infrared thermal imagers, acoustic fingerprint acquisition arrays, lidar, environmental monitoring sensors, exhaust fan controllers, lighting dimming modules, fire sprinkler solenoid valves, and intelligent inspection robots.

[0009] The edge processing layer is used to perform edge-side task segmentation and preliminary intelligent analysis. The edge processing layer includes multiple sets of edge computing nodes deployed in the substation terminal equipment room. Each edge computing node is equipped with a computing power processing unit consisting of a multi-core central processing unit and a dedicated artificial intelligence acceleration chip, and is also equipped with a solid-state drive storage array and redundant power supply.

[0010] The network transport layer is used to realize the logical connection between the edge processing layer and the cloud decision layer. The network transport layer adopts industrial private network slicing technology based on predetermined generations of mobile communication technology and is configured with an industrial gateway that supports multi-carrier link backup.

[0011] The cloud-based decision layer is used to realize dynamic resource scheduling between the cloud and the edge, multi-source information fusion, global status diagnosis, and generation of closed-loop auxiliary control commands. The cloud-based decision layer includes a data storage system, a model training server, a digital twin engine, and a scheduling logic processing unit.

[0012] This invention also provides a cloud-edge collaborative intelligent inspection and auxiliary control scheduling method for substations, implemented based on the aforementioned scheduling system, comprising the following steps:

[0013] S1 utilizes a multi-dimensional sensing network to collect substation operation data in real time, and performs preliminary formatting and time synchronization on the raw data to form raw data frames with unified time stamps.

[0014] S2 utilizes edge computing nodes to perform real-time filtering and feature extraction on the collected data, while simultaneously dividing the auxiliary control commands into tasks and sending the processed structured data and task requests to the cloud decision layer.

[0015] S3: The cloud decision layer dynamically allocates computing tasks between the cloud and the edge based on the current network bandwidth status, the computing load of the edge nodes, and the priority of the inspection tasks, and synchronously updates the algorithm model parameters of the edge.

[0016] S4, the cloud-based decision-making layer gathers multi-dimensional data from multiple edge nodes, combines historical operation and maintenance databases with expert systems, and uses long short-term memory neural networks to predict and analyze the operating trends of equipment.

[0017] S5, based on the predictive analysis results, combined with the substation topology, performs correlated fault diagnosis and constructs a digital twin model of the substation;

[0018] S6 maps real-time monitoring data to a digital twin model, calculates the state deviation between the digital twin model and the physical entity, and generates a global inspection report and scheduling decision recommendations.

[0019] S7: The system automatically issues control commands to the auxiliary control equipment at the substation site based on the scheduling decision suggestions, realizes the logical linkage between inspection tasks and auxiliary control actions, and monitors the execution results in real time through a feedback mechanism.

[0020] Furthermore, in this invention, the substation operation data includes circuit breaker positions, disconnector status, and instrument readings collected by industrial-grade high-definition cameras; abnormal heating of electrical connection points detected by infrared thermal imagers; electromagnetic noise characteristics of transformers and reactors identified by acoustic signature acquisition arrays; a three-dimensional spatial map of the substation constructed by lidar to locate the path planning coordinates of the inspection robot; ambient temperature, humidity, sulfur hexafluoride gas concentration, and smoke concentration monitored by environmental monitoring sensors; and all collected analog signals are converted into digital signals in binary two's complement form through analog-to-digital conversion circuits.

[0021] Furthermore, in step S3, the scheduling algorithm determines the execution location of the computational task by minimizing a cost function composed of processing latency and system energy consumption; wherein the expression for the cost function is:

[0022]

[0023] In the formula, Indicates the total cost. The estimated time required to process tasks at the edge. The round-trip delay of data at the transport layer. The computation time required to process deep models in the cloud. The energy consumption for edge node operation. , , , These are the corresponding weighting coefficients, and all weighting coefficients are dynamically adjusted based on the current operating conditions of the substation.

[0024] Furthermore, in S2, the task segmentation process is as follows: preprocessing tasks with computational load within a preset computing power range and real-time requirements higher than a preset latency threshold are retained for execution at the edge, while large-scale computing tasks involving global optimization or deep model training are migrated to the cloud decision layer through the network transmission layer.

[0025] Furthermore, in S3, the deep reinforcement learning scheduling algorithm adopts an actor-critic framework, taking the processor utilization of the edge computing node, the current available network bandwidth, and the urgency of the inspection task as state inputs, and the task offloading ratio as action outputs, and maximizing the cumulative reward function to configure computing resources.

[0026] Furthermore, in S4, the Long Short-Term Memory Neural Network consists of an input layer, a hidden layer containing a forget gate, an input gate, an output gate, and an output layer; the input layer receives serialized data from substation equipment within a preset number of sampling periods, the serialized data including voltage, current, temperature, and voiceprint feature values; the output layer calculates the probability distribution of equipment operating deviations within a predetermined time period, and the Long Short-Term Memory Neural Network captures long-term dependencies in the data through a gating mechanism to predict equipment operating trends.

[0027] Furthermore, in S5, the associated fault diagnosis obtains the electrical topology connection relationship of the substation. When a circuit breaker trip signal is detected, the scheduling logic processing unit associates the real-time monitoring data of the transformer, bus and instrument transformer connected to the circuit breaker, and determines the fault point and its impact range by comparing multi-source information.

[0028] Furthermore, in S5, the digital twin model construction process is as follows: using the three-dimensional point cloud data acquired by the lidar, a surface mesh model is generated through the Poisson reconstruction algorithm, and the real-time running data collected by the perception execution layer is bound to the corresponding node of the mesh model; the calculation of the state deviation adopts the Euclidean distance algorithm, comparing the theoretical running curve in the digital twin model with the measured running curve of the physical entity, and when the deviation exceeds the preset deviation threshold, the system determines that the device is in a sub-healthy state and generates an early warning report.

[0029] Furthermore, in S7, the closed-loop auxiliary control commands include start / stop control of the exhaust fan controller, brightness adjustment of the lighting dimming module, and linkage control of the fire sprinkler solenoid valve; after issuing the command, the system verifies the action status of the actuator in real time through the current change signal or displacement sensor signal fed back by the sensing and execution layer, forming a control closed loop; when the battery power of the intelligent inspection robot is lower than the preset power threshold, the intelligent inspection robot performs path planning and returns to the charging position for recharging, and the status data during the charging process is synchronized to the cloud decision layer.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] (1) By constructing a four-layer collaborative architecture consisting of a perception execution layer, an edge processing layer, a network transmission layer, and a cloud decision layer, this invention breaks the architectural limitations of traditional inspection systems that rely excessively on full cloud computing. It pushes preprocessing tasks with low latency requirements down to the edge side for execution, significantly reducing the bandwidth load of network transmission and the instruction response latency, and can fully meet the real-time auxiliary control requirements of substations in emergency situations.

[0032] (2) The present invention adopts a cloud-edge task dynamic scheduling mechanism based on deep reinforcement learning, which can dynamically divide the computing tasks according to the real-time computing power load, network bandwidth status and inspection task priority of the edge node, realize the flexible allocation of edge computing power and the global optimal configuration of cross-domain resources, effectively avoid the performance bottleneck of edge nodes when processing high-density computing tasks, and improve the system's adaptability to different operating conditions.

[0033] (3) This invention achieves full-process logical linkage of inspection data collection, status diagnosis and analysis and auxiliary control command execution through the deep coupling of real-time synchronization of cloud edge data and command feedback mechanism. It constructs a complete response link from abnormal perception to closed-loop handling, solves the problem of logical disconnect between inspection tasks and auxiliary control actions in traditional solutions, and greatly improves the efficiency and accuracy of substation abnormal handling.

[0034] (4) This invention integrates multi-dimensional sensing data and digital twin technology. Through correlation fault diagnosis and equipment operation trend prediction, it can realize full-dimensional visualization control and predictive operation and maintenance of substation equipment status. At the same time, combined with the multi-link redundant industrial private network transmission scheme, it effectively improves the system's operational robustness under complex electromagnetic environment and variable communication conditions, providing reliable technical support for the implementation of unmanned operation and maintenance of substations. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the system structure of the present invention.

[0036] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0038] like Figure 1As shown, this invention discloses a cloud-edge collaborative intelligent substation inspection and control scheduling system. At the physical entity level, this system is divided into a perception and execution layer, an edge processing layer, a network transmission layer, and a cloud decision-making layer. The perception and execution layer, acting as the system's nerve endings, is directly deployed in the high-voltage equipment area, control room, and surrounding environment of the substation. The core components of this layer include an industrial-grade high-definition camera, which integrates a back-illuminated complementary metal-oxide-semiconductor (CMOS) photosensitive element, possessing 4K ultra-high-definition resolution. The camera achieves 360° horizontal and 180° vertical omnidirectional rotation via an industrial-grade pan-tilt unit, with a positioning accuracy of 0.05°, ensuring precise capture of circuit breaker opening / closing indicators, physical contact gaps of disconnecting switches, and real-time readings of various analog pointer instruments. The infrared thermal imager uses an uncooled focal plane array detector, capable of sensing infrared radiation in the 8-14μm band, with a thermal sensitivity better than 0.05℃, specifically designed to detect localized overheating at electrical connection points caused by excessive contact resistance. The acoustic signature acquisition array consists of multiple sets of high-sensitivity condenser microphones, which are arranged in a specific spatial geometry to form a microphone array. Beamforming technology is used to eliminate environmental wind noise and background noise, focusing on extracting the specific frequency band acoustic wave characteristics generated by the vibration of the transformer core.

[0039] In addition, the perception and execution layer is equipped with an intelligent inspection robot. This robot's chassis uses a four-wheel independent drive and suspension system, enabling it to traverse small cable trenches within the substation. The lidar mounted on top of the robot has 16 or 32-line scanning capability, emitting and receiving hundreds of thousands of laser pulses per second, constructing a 3D point cloud map of the surrounding environment by measuring the pulse round-trip time. The environmental monitoring sensor group includes an electrochemical SF6 gas sensor, a semiconductor smoke detector, and a high-precision integrated temperature and humidity transmitter. At the execution level, this layer integrates exhaust fan controllers, lighting dimming modules, and fire sprinkler solenoid valves. These actuators are connected to edge computing nodes via industrial relays or pulse-width modulation signals, receiving control commands from the upper layer.

[0040] The edge processing layer, serving as the central computing hub within the substation, is deployed in the secondary equipment room. Its core hardware consists of multiple high-performance edge computing nodes, each employing a heterogeneous computing architecture. Specifically, a multi-core central processing unit (CPU) handles general logic, communication protocol conversion, and file system management; a dedicated AI acceleration chip boasts up to 200 TOPS of deep learning inference computing power, specifically designed for running object detection, image segmentation, and voiceprint recognition algorithms. Each node also features a storage array composed of multiple solid-state drives (SSDs), utilizing redundant disk array technology to ensure data security. To cope with the complex electromagnetic interference environment of the substation, the edge computing node's power module employs a dual-redundant design, supporting both AC 220V and DC 110V inputs, and possesses high-level surge protection and electromagnetic compatibility performance.

[0041] The network transport layer constructs a digital tunnel connecting the substation and the cloud. This embodiment employs an industrial private network slicing scheme based on 5G technology. By deploying 5G micro base stations within the substation and utilizing network function virtualization technology, logical slices with extremely high priority and guaranteed bandwidth are allocated for inspection services. This slice employs orthogonal frequency division multiplexing (OFDM) technology at the physical layer and implements a dedicated resource block allocation strategy at the link layer, ensuring that the end-to-end transmission latency of inspection video streams and control commands is strictly controlled within 20ms. The industrial gateway configured on-site has multi-link load balancing capabilities, enabling simultaneous access to the 5G networks of multiple operators such as China Mobile and China Telecom. When the primary link signal is interfered with, causing a packet loss rate exceeding 1%, the gateway uses a lossless switching algorithm to redirect data packets to a backup link within 10ms.

[0042] The cloud-based decision-making layer is deployed in the power company's central computer room or on a dedicated cloud platform. Its core component, the data storage system, adopts a distributed object storage architecture, enabling unified storage of massive amounts of unstructured data (such as video, audio, and point clouds) and structured data (such as telemetry, remote signaling, and inspection logs). The model training server cluster utilizes high-performance graphics processors for offline training and incremental learning of deep neural networks. The digital twin engine, based on Unreal Engine or similar 3D rendering technology, constructs a digital twin of the substation. The scheduling logic processing unit, as the highest decision-making body, is responsible for running the global resource scheduling algorithm and closed-loop auxiliary control logic.

[0043] like Figure 2 As shown, the scheduling method of the above system is as follows:

[0044] First, the multi-dimensional sensing network initiates real-time data acquisition. Video streams captured by industrial-grade high-definition cameras, temperature matrices generated by infrared thermal imagers, audio waveforms recorded by acoustic wave arrays, and analog signals from environmental sensors are all digitized by the acquisition units of the sensing layer. Specifically, the analog signals are converted into binary two's complement digital signals via a 16-bit high-precision analog-to-digital converter. To ensure time alignment of data from different sensors, the system employs a precise time protocol based on the BeiDou Navigation Satellite System, embedding microsecond-level time stamps in the header of each data frame. These raw data frames with unified time stamps are then aggregated to the edge computing nodes via an on-site industrial Ethernet network.

[0045] Edge computing nodes filter incoming raw data in real time. AI acceleration chips run pre-defined lightweight convolutional neural networks to filter video frames, removing redundant images with unchanged backgrounds. Simultaneously, edge nodes perform Fast Fourier Transform on the voiceprint signal to extract feature frequency distributions. For auxiliary control commands, edge nodes perform logical validity preprocessing, such as determining whether the exhaust fan start command conflicts with the current physical feedback state of the fan. The processed structured feature data (such as instrument readings, coordinates of abnormal heat points, and voiceprint feature vectors) and specific computational task requests are encapsulated and sent to the cloud-based decision layer.

[0046] Subsequently, the cloud-based decision-making layer monitors in real time the bandwidth utilization of the current 5G network, the CPU occupancy rate of each edge node, and the priority of pending inspection tasks. The system utilizes a deep reinforcement learning-based scheduling algorithm to dynamically allocate computing load between the cloud and the edge. The scheduling algorithm determines the optimal execution location of computing tasks by minimizing a cost function comprised of processing latency and system energy consumption. The mathematical expression for the cost function C is as follows:

[0047]

[0048] In this formula, C represents the total scheduling cost. This indicates the estimated time required to process the task at the edge, which depends on the current available computing power of the edge node. This indicates the time delay for data to travel back and forth between the two ends at the network transport layer. This indicates the computation time required for a cloud server to process a complex deep learning model. e represents the additional energy consumption incurred by the edge node in performing this task. , , , These are the corresponding weighting coefficients. The system dynamically adjusts these weighting coefficients based on the current operating conditions of the substation. For example, during thunderstorms or when equipment triggers an emergency alarm, the system automatically increases these weighting coefficients. and This sacrifices energy consumption for minimal processing latency; however, during routine inspections, the latency is increased. To achieve energy-efficient operation, deep reinforcement learning networks continuously observe the state space (including load, bandwidth, and task size) and execute actions (determining the offloading ratio). They evolve based on the negative value of the cost function as a reward signal, thereby achieving optimal allocation of cross-domain resources.

[0049] After receiving multi-source data aggregated from various edge nodes, the cloud-based decision layer activates a Long Short-Term Memory (LSTM) neural network for trend prediction. The LTM neural network's input layer receives serialized data from the past 128 sampling periods, including historical temperature curves of the equipment, load current fluctuations, and trends in acoustic signature characteristics. Through the logical collaboration of forget gates, input gates, and output gates in the hidden layer, the network can capture long-term dependencies in the data. The output layer calculates the probability distribution of equipment operational deviations or failures within the next 24 hours. For example, when the system detects a gradual increase in high-frequency components in the acoustic signature of a transformer, accompanied by a slow rise in bushing temperature under infrared thermography, the LTM neural network will determine that the equipment exhibits an internal discharge trend.

[0050] Based on the above predictive analysis results, the system performs correlation fault diagnosis by combining the electrical topology of the substation. The dispatch logic processing unit retrieves the single-line diagram logic of the substation. When an abnormal signal is detected in a circuit breaker, it automatically retrieves all real-time monitoring indicators of the connected instrument transformers, surge arresters, and busbars. Through cross-validation of multi-source information, false alarm interference from sensors is eliminated, and the physical fault point is accurately located. Simultaneously, the digital twin engine updates the three-dimensional geometric model of the substation based on the latest point cloud data collected by LiDAR, constructing a digital twin model that is highly consistent with the physical world.

[0051] Then, the system maps all operational parameters collected in real time by the sensing layer into the digital twin model. By simulating the operational behavior of the physical entity in virtual space, the system calculates the state deviation between the theoretical state value predicted by the digital twin model and the measured value of the physical entity. The system first uses a physical dynamics simulation module, combined with the current load current, ambient temperature, and operating time, to calculate the theoretical temperature distribution curve T of the device under ideal conditions. theory (t). Simultaneously, the sensing layer acquires the measured temperature curve T. actual (t). The system employs a dynamic time warping algorithm within a sliding window to calculate the similarity distance between two curves. The deviation D is calculated using the weighted Euclidean distance formula:

[0052]

[0053] In the formula, These are the weighting coefficients for different sampling points, with points closer to the current time having higher weights. When the D value continuously exceeds the preset alarm threshold, the system will initiate deep diagnostic logic. For example, if the measured temperature is much higher than the theoretical simulated temperature, and the acoustic signature indicates increased mechanical vibration, the system will infer that there may be localized overheating inside the transformer due to loose windings.

[0054] Finally, based on scheduling decision recommendations, the system automatically issues closed-loop control commands. For example, when the SF6 concentration in the main transformer room exceeds the standard, the cloud-based decision layer issues a command to the edge node, which then forces the exhaust fan to start via a relay output signal. During execution, sensors at the sensing and execution layer monitor the fan's current changes and wind speed feedback in real time, forming a complete control closed loop. If the fan fails to start, the system immediately pushes an emergency alarm to the mobile terminal of maintenance personnel via the 5G private network and triggers the lighting dimming module to switch to emergency indication mode.

[0055] Through the deep integration of the above system architecture and methodology, this embodiment realizes intelligent operation of the entire substation inspection process, from "perception" to "decision-making" and then to "control".

[0056] In summary, this invention constructs an intelligent inspection and control scheduling system with self-learning, self-diagnosis, and self-healing capabilities through deep collaboration among the perception and execution layer, edge processing layer, network transmission layer, and cloud decision-making layer. This system not only significantly improves the real-time performance and accuracy of substation inspections but also effectively resolves the contradiction between massive data processing and communication bandwidth limitations through a cloud-edge collaborative computing power scheduling mechanism. The introduction of digital twins enables substation operation and maintenance management to move from macro-level digitization to micro-level precision, laying a solid technical foundation for achieving safe and reliable operation and unmanned operation and maintenance of power systems.

[0057] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A cloud-edge collaborative intelligent substation inspection auxiliary control and dispatch system, characterized in that, include: The perception and execution layer is used to construct a multi-dimensional perception network and collect substation operation data in real time. The perception and execution layer is deployed at the substation site and includes industrial-grade high-definition cameras, infrared thermal imagers, acoustic fingerprint acquisition arrays, lidar, environmental monitoring sensors, exhaust fan controllers, lighting dimming modules, fire sprinkler solenoid valves, and intelligent inspection robots. The edge processing layer is used to perform edge-side task segmentation and preliminary intelligent analysis. The edge processing layer includes multiple sets of edge computing nodes deployed in the substation terminal equipment room. Each edge computing node is equipped with a computing power processing unit consisting of a multi-core central processing unit and a dedicated artificial intelligence acceleration chip, and is also equipped with a solid-state drive storage array and redundant power supply. The network transport layer is used to realize the logical connection between the edge processing layer and the cloud decision layer. The network transport layer adopts industrial private network slicing technology based on predetermined generations of mobile communication technology and is configured with an industrial gateway that supports multi-carrier link backup. The cloud-based decision layer is used to realize dynamic resource scheduling between the cloud and the edge, multi-source information fusion, global status diagnosis, and generation of closed-loop auxiliary control commands. The cloud-based decision layer includes a data storage system, a model training server, a digital twin engine, and a scheduling logic processing unit.

2. A cloud-edge collaborative intelligent inspection and control scheduling method for substations, characterized in that, This is based on the cloud-edge collaborative intelligent substation inspection auxiliary control and dispatch system described in claim 1. Includes the following steps: S1 utilizes a multi-dimensional sensing network to collect substation operation data in real time, and performs preliminary formatting and time synchronization on the raw data to form raw data frames with unified time stamps. S2 utilizes edge computing nodes to perform real-time filtering and feature extraction on the collected data, while simultaneously dividing the auxiliary control commands into tasks and sending the processed structured data and task requests to the cloud decision layer. S3: The cloud decision layer dynamically allocates computing tasks between the cloud and the edge based on the current network bandwidth status, the computing load of the edge nodes, and the priority of the inspection tasks, and synchronously updates the algorithm model parameters of the edge. S4, the cloud-based decision-making layer gathers multi-dimensional data from multiple edge nodes, combines historical operation and maintenance databases with expert systems, and uses long short-term memory neural networks to predict and analyze the operating trends of equipment. S5, based on the predictive analysis results, combined with the substation topology, performs correlated fault diagnosis and constructs a digital twin model of the substation; S6 maps real-time monitoring data to a digital twin model, calculates the state deviation between the digital twin model and the physical entity, and generates a global inspection report and scheduling decision recommendations. S7: The system automatically issues control commands to the auxiliary control equipment at the substation site based on the scheduling decision suggestions, realizes the logical linkage between inspection tasks and auxiliary control actions, and monitors the execution results in real time through a feedback mechanism.

3. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 2, characterized in that, The substation operation data includes circuit breaker locations, disconnector status, and instrument readings collected by industrial-grade high-definition cameras; abnormal heating of electrical connection points detected by infrared thermal imagers; electromagnetic noise characteristics of transformers and reactors identified by acoustic signature arrays; a three-dimensional spatial map of the substation constructed by lidar, which provides the path planning coordinates for the inspection robot; ambient temperature, humidity, sulfur hexafluoride gas concentration, and smoke concentration monitored by environmental monitoring sensors; and all collected analog signals are converted into digital signals in binary two's complement form through analog-to-digital conversion circuits.

4. The cloud-edge collaborative intelligent substation inspection auxiliary control and scheduling method according to claim 3, characterized in that, In step S3, the scheduling algorithm determines the execution location of the computational task by minimizing a cost function composed of processing latency and system energy consumption; wherein the expression for the cost function is: In the formula, Indicates the total cost. The estimated time required to process tasks at the edge. The round-trip delay of data at the transport layer. The computation time required to process deep models in the cloud. The energy consumption for edge node operation. , , , These are the corresponding weighting coefficients, and all weighting coefficients are dynamically adjusted based on the current operating conditions of the substation.

5. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 4, characterized in that, In S2, the task segmentation process is as follows: preprocessing tasks with computational load within a preset computing power range and real-time requirements higher than a preset latency threshold are retained and executed at the edge, while large-scale computing tasks involving global optimization or deep model training are migrated to the cloud decision layer through the network transmission layer.

6. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 5, characterized in that, In S3, the deep reinforcement learning scheduling algorithm adopts the actor critic framework, taking the processor utilization of the edge computing node, the current available network bandwidth, and the urgency of the inspection task as state inputs, and the task offloading ratio as action outputs, and maximizing the cumulative reward function to configure computing resources.

7. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 6, characterized in that, In S4, the Long Short-Term Memory Neural Network consists of an input layer, a hidden layer containing a forget gate, an input gate, an output gate, and an output layer. The input layer receives serialized data from substation equipment within a preset number of sampling periods. The serialized data includes voltage, current, temperature, and acoustic signature features. The output layer calculates the probability distribution of equipment operation deviations within a predetermined time period. The Long Short-Term Memory Neural Network captures long-term dependencies in the data through a gating mechanism and performs predictions of equipment operation trends.

8. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 7, characterized in that, In S5, the correlation fault diagnosis obtains the electrical topology connection relationship of the substation. When a circuit breaker trip signal is detected, the scheduling logic processing unit associates the real-time monitoring data of the transformer, bus and instrument transformer connected to the circuit breaker, and determines the fault point and its impact range by comparing multi-source information.

9. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 8, characterized in that, In S5, the digital twin model construction process is as follows: using the three-dimensional point cloud data acquired by the lidar, a surface mesh model is generated through the Poisson reconstruction algorithm, and the real-time running data collected by the perception execution layer is bound to the corresponding node of the mesh model; the state deviation is calculated using the Euclidean distance algorithm, comparing the theoretical running curve in the digital twin model with the measured running curve of the physical entity, and when the deviation exceeds the preset deviation threshold, the system determines that the device is in a sub-healthy state and generates an early warning report.

10. The cloud-edge collaborative substation intelligent inspection auxiliary control and scheduling method according to claim 9, characterized in that, In S7, the closed-loop auxiliary control commands include start / stop control of the exhaust fan controller, brightness adjustment of the lighting dimming module, and linkage control of the fire sprinkler solenoid valve; after the system issues the command, it verifies the action status of the actuator in real time through the current change signal or displacement sensor signal fed back by the sensing and execution layer, thus forming a control closed loop; When the battery level of the intelligent inspection robot is lower than a preset power threshold, the intelligent inspection robot performs path planning and returns to the charging location to recharge. The status data during the charging process is synchronized to the cloud decision layer.

Images