Intelligent inspection and fault prediction system based on computing power service

Abstract

The application discloses a smart inspection and fault prediction system based on computing power service and belongs to the technical field of power grid inspection, and specifically comprises the following steps: collecting substation environment data through a sensor array; analyzing power grid dispatching instructions through an instruction analysis module to determine target equipment and an expected state path; constructing an equipment operation intention topology network through a topology network construction module and marking key nodes; generating a perception set containing equipment state recognition and robot movement control tasks according to environment data through a task screening module, and screening tasks outputting key node state parameters as a core task set; calculating the minimum computing power of core tasks according to key node weights and environment interference coefficients through a computing power allocation module, and triggering non-core task degradation when the computing power is insufficient; and inputting equipment state parameters output by the core tasks into a prediction network through a fault prediction module to generate an abnormal coefficient and a maintenance sequence; the application guarantees the real-time performance of key tasks through dynamic allocation of computing power resources, and improves the power grid operation and maintenance efficiency and reliability.

Description

Technical Field

[0001] This invention relates to the field of power grid inspection technology, and more specifically to an intelligent inspection and fault prediction system based on computing power services. Background Technology

[0002] With the advancement of new power system construction, substations, as core hubs of the power grid, are experiencing increasingly complex equipment types, including GIS switchgear and smart circuit breakers. Their operating environment is affected by multiple factors such as electromagnetic interference, terrain undulations, and climate fluctuations, placing higher demands on the real-time performance and accuracy of equipment status monitoring. Intelligent inspection and fault prediction, as key technologies for ensuring the safe operation of the power grid, achieve dynamic perception of equipment status and fault early warning through sensor acquisition, data processing, and intelligent analysis, becoming an important support for the digital transformation of the power grid. Simultaneously, the widespread adoption of computing services, such as edge computing and distributed computing power scheduling, provides the foundation for the real-time processing of massive inspection data, driving the upgrade of inspection models from periodic inspections to precise perception and predictive maintenance, significantly improving the efficiency of power grid operation and maintenance. In existing technologies, intelligent inspection systems typically employ sensor arrays to collect data such as equipment images and environmental parameters. These are combined with deep learning models to achieve equipment status recognition, including determining the open / closed status of disconnectors and detecting insulator defects. Furthermore, fault prediction algorithms, such as LSTM and CNN, are used to generate equipment health assessments. Some systems introduce edge computing nodes to reduce data transmission latency and allocate computing resources through task scheduling mechanisms, thus meeting the needs of routine inspection scenarios to a certain extent. For example, by pre-prioritizing tasks, higher computing power is allocated to equipment image recognition tasks to ensure recognition accuracy; and by combining historical fault data to optimize prediction model parameters, the timeliness of fault warnings is improved. However, existing technologies lack targeted task selection and computing power allocation, failing to clearly distinguish between core tasks (directly related to critical node status verification) and non-core tasks. This can easily lead to non-core tasks consuming excessive computing power, causing delays in critical tasks such as switch status identification due to insufficient computing power. Furthermore, the computing power adjustment mechanism lacks dynamic adaptability. When environmental interference, such as strong electromagnetic fields or complex terrain, increases the computing power demand of core tasks, it cannot quickly downgrade non-core tasks to ensure the computing power supply for core tasks, affecting the overall efficiency of inspection and fault prediction. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent inspection and fault prediction system based on computing power services, and to solve the following technical problems: In existing technologies, task selection and computing power allocation lack specificity and fail to clearly distinguish between core tasks (i.e., directly related to critical node status verification) and non-core tasks. This can easily lead to non-core tasks consuming too much computing power, causing critical tasks to be delayed due to insufficient computing power.

[0004] The objective of this invention can be achieved through the following technical solutions: The intelligent inspection and fault prediction system based on computing power services includes: A sensor array is used to collect environmental data of the substation, including equipment image information, terrain geometry information, and electromagnetic field strength information. The instruction parsing module is used to parse the equipment operation instructions issued by the power grid dispatching system and determine the target operation equipment and its expected state change path. The topology construction module is used to construct a topology network for device operation intentions. The nodes of the topology network represent the physical states that the target operating device needs to verify, and the connecting lines between nodes represent the dependencies of state verification. Nodes located at the beginning of the power grid fault propagation chain and whose position weight exceeds the protection threshold are marked as critical nodes. The task filtering module is used to generate a set of perception tasks based on the type of environmental data, including equipment status recognition tasks and robot movement control tasks. It filters out tasks in the perception task set that directly output the physical state parameters of key nodes and marks them as the core task set. The computing power allocation module is used to calculate the minimum computing power guarantee value for each task in the core task set based on the location weight of key nodes and the environmental interference coefficient. When the actual computing power occupied by the core task set is lower than the sum of the minimum computing power guarantee values, the non-core task computing function is degraded. The fault prediction module is used to input the equipment status parameters output by the core task set into the fault prediction network to generate equipment status anomaly coefficients and maintenance operation sequences.

[0005] As a further aspect of the present invention: the process of parsing equipment operation instructions issued by the power grid dispatching system in the instruction parsing module is as follows: It receives standardized operation command streams issued by the power grid dispatching system. The command streams conform to the power system operation protocol specifications. It parses the command text structure and extracts the pairing relationship between operation verbs and equipment identifiers. The operation verbs include tripping operation, closing operation, and voltage regulation operation types. The system queries the power grid topology database based on the device identifier to obtain the physical installation location of the device and its upstream and downstream relationships in the electrical connection diagram; it derives the expected state change path based on the operation verb type, such as the opening and closing angle change path corresponding to the tripping operation and the voltage level change path corresponding to the voltage regulation operation, both of which include the state change direction and time constraints. When multiple device operation commands have execution order dependencies, a time-series constraint relationship chain for state changes is established to ensure that the dependency relationship of the device operation intention topology accurately reflects the operation sequence logic.

[0006] As a further aspect of the present invention: the specific process of defining key nodes in the topology network construction module is as follows: A power grid fault propagation chain model is constructed, which represents the path relationship of fault propagation from the initiating device to the associated devices. The weight value of the topology node is calculated, and the weight value is proportional to the number of downstream devices affected by the node in the fault propagation chain. Locate the starting node of the fault propagation chain, which is defined as a source node with no upstream dependency; mark the nodes that are simultaneously located at the starting point and whose position weight value exceeds the system protection threshold as critical nodes, and update the set of critical nodes when the power grid topology changes.

[0007] As a further aspect of the present invention: the filtering process for the core task set in the task filtering module is as follows: In the device operation intention topology network, identify the direct dependent nodes of key nodes to form a set of key verification targets; analyze the data processing targets of each task in the perception task set, and select the tasks that can output the physical state parameters of the key verification target set. Calculate the sum of the minimum computing power guarantee values ​​of the selected tasks; compare the sum with the real-time available computing power of the edge computing unit; when the sum exceeds the available computing power, remove tasks in descending order of location weight value until the computing power constraint is met; take the tasks that are not removed as the core task set, and turn the removed tasks into background monitoring tasks and set the maximum computing power quota.

[0008] As a further aspect of the present invention: in the computing power allocation module, the process of calculating the minimum computing power guarantee value for each task in the core task set is as follows: Mark the power grid physical equipment corresponding to the core task set as associated equipment, and calculate the basic computing power requirement of the task based on the voltage level coefficient and historical failure rate of the associated equipment. The terrain geometry information is analyzed to determine the rate of change of surface curvature. When the rate of change exceeds a safety threshold, a terrain interference component is generated. The power frequency noise energy density is extracted from the electromagnetic field strength information, and an electromagnetic interference component is generated based on the energy density distribution. The terrain interference component and the electromagnetic interference component are then fused to form an environmental interference intensity coefficient. Multiply the basic computing power requirement of the task by the environmental interference intensity coefficient, and output the product as the minimum computing power guarantee value.

[0009] As a further aspect of the present invention: in the computing power allocation module, the generation process of the environmental interference intensity coefficient is as follows: A linear proportional function is established between the rate of change of surface curvature and the terrain interference component, and an exponential growth function is established between the power frequency noise energy density and the electromagnetic interference component. The two types of interference components are fused according to the preset regional sensitivity weights, and the fusion result is the environmental interference intensity coefficient.

[0010] As a further aspect of the present invention: the specific process of triggering the degradation of non-core task computing functions in the computing power allocation module is as follows: When the actual computing power occupied by the core task set is lower than the total guaranteed value, the terrain conditions of the robot's mobile control task are detected. If the standard deviation of terrain undulation is lower than the threshold and the obstacle density is sparse, real-time 3D point cloud reconstruction is turned off and historical path tracking is activated. The system detects whether the image acquisition task associated with the detection equipment is outside the set of key verification targets. If it is outside the set, the continuous video analysis is converted into key frame sampling analysis and the sampling interval is extended. During the degradation process, the original sensor data is continuously cached. When the total remaining computing power of the edge computing unit is continuously higher than the recovery threshold, the cached data is supplemented by calculation.

[0011] As a further aspect of the present invention: the process of generating the equipment state anomaly coefficient in the fault prediction module is as follows: Obtain the real-time status parameters of the core task set monitoring equipment, extract the expected status standard values ​​of the corresponding equipment from the equipment operation intention topology network, calculate the absolute deviation between the real-time parameters and the standard values, and construct the deviation time series of continuous inspection cycles. A sliding window algorithm is used to identify monotonically increasing trends in time series. When the cumulative increase of the increasing trend exceeds the equipment design tolerance range, the anomaly coefficient calculation is initiated. The anomaly coefficient of non-linear growth is calculated based on the trend duration and the rate of increase.

[0012] As a further aspect of the present invention: in the fault prediction module, the process of generating the maintenance operation sequence is as follows: When the abnormal coefficient exceeds the operation threshold, locate the direct upstream device of the device corresponding to the abnormal coefficient in the topology network, and determine whether the position weight value of the upstream device exceeds the preset ratio of the system protection threshold. If the weight value exceeds the preset proportion of the system protection threshold and the abnormal coefficient growth rate is higher than the transmission threshold, a device isolation and backup switching instruction is generated; if the upstream device is redundantly configured, the maximum delayed maintenance time window is calculated based on the health score.

[0013] The beneficial effects of this invention are: This invention constructs a topology network for device operation intentions using a topology network construction module. Based on the power grid fault propagation chain markers indicating key nodes at the starting end with position weights exceeding protection thresholds, a task selection module selects core tasks that directly output the physical state parameters of key nodes, clearly distinguishing between core and non-core tasks. This addresses the lack of specificity in task selection and computing power allocation in existing technologies. The computing power allocation module calculates the minimum computing power guarantee value for core tasks based on the position weights of key nodes and environmental interference coefficients. When the actual computing power occupied by core tasks falls below the total guarantee value, the computing function of non-core tasks is downgraded, achieving dynamic adaptation of computing power and preventing non-core tasks from consuming too much computing power, which could lead to delays in key tasks due to insufficient computing power. Simultaneously, the fault prediction module, combining the dependencies of device status verification in the topology network, uses the physical state parameters output by core tasks to generate device status anomaly coefficients and maintenance operation sequences. This fully considers the impact of related devices, improving the accuracy of fault early warning, effectively addressing scenarios where environmental interference increases the computing power demand of core tasks, ensuring the real-time performance of key tasks, and improving the overall efficiency of inspection and fault prediction. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of the modules of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, the present invention is an intelligent inspection and fault prediction system based on computing power services, comprising: The sensor array is deployed inside and around the substation to collect comprehensive environmental data. This includes equipment image information covering the appearance and operating status of various power equipment, terrain geometry information recording the surface morphology and equipment distribution within the inspection area, and electromagnetic field strength information capturing real-time changes in the surrounding electromagnetic field. All of this data provides fundamental support for subsequent status analysis and task scheduling.

[0018] The instruction parsing module interfaces with the power grid dispatching system, responsible for receiving and parsing equipment operation instructions issued by the system. By breaking down the instruction content layer by layer, it accurately identifies the target equipment to be operated and outlines the expected state change path of the equipment during operation, providing a clear reference standard for subsequent equipment state verification.

[0019] The topology construction module builds a topology network of the device's operational intent based on the expected state change path of the target device. In this topology network, each node corresponds to a physical state that the target device needs to verify, and the connections between nodes visually represent the dependencies between different state verifications. Simultaneously, combined with the analysis results of the power grid fault propagation chain, nodes at the beginning of the propagation chain and whose position weight reaches the protection threshold are marked as critical nodes, highlighting the device states that require key monitoring.

[0020] Based on the environmental data collected by the sensor array, the task selection module generates a set of perception tasks containing multiple tasks. These include equipment status identification tasks that identify the operating status of equipment, as well as robot movement control tasks that control the movement of the inspection robot. On this basis, tasks that can directly output the physical state parameters of key nodes are selected from the perception task set and marked as the core task set, thus identifying the critical tasks that need to be prioritized.

[0021] The computing power allocation module combines the location weights of key nodes with an environmental interference coefficient derived from terrain geometry and electromagnetic field strength information to calculate the minimum computing power guarantee value for each task in the core task set to ensure its normal operation. When the actual computing power used by the core task set is lower than the sum of these minimum computing power guarantee values, the computing functions of non-core tasks are automatically degraded to prioritize the computing power needs of the core tasks.

[0022] The fault prediction module inputs the equipment status parameters output from the core task set into a preset fault prediction network. Through the analysis and processing of these key data, it generates an equipment status anomaly coefficient that reflects the degree of equipment malfunction and simultaneously generates a targeted maintenance operation sequence to provide specific action guidance for power grid operation and maintenance.

[0023] In a preferred embodiment of the present invention, the process of parsing the equipment operation instructions issued by the power grid dispatching system in the instruction parsing module is as follows: First, the module receives standardized operation command streams from the power grid dispatching system. These command streams strictly adhere to power system operation protocol specifications, covering command encoding formats, data field definitions, and verification rules to ensure consistency in command transmission across different systems. During the parsing phase, the module uses semantic analysis technology to deconstruct the command text structure. By identifying key segments in the commands, it separates the pairing relationships between operation verbs and equipment identifiers. Specifically, operation verbs include opening, closing, and voltage regulation operations, while equipment identifiers use a unified encoding rule, containing information such as equipment type, assigned bay, and unique number.

[0024] After extracting the pairing relationships, the module accesses the power grid topology database based on the device identifier. This database stores details of the physical installation locations of all equipment in the station, including specific coordinates, installation height, and surrounding environmental characteristics. Simultaneously, it retrieves the upstream and downstream relationships of the equipment in the electrical connection diagram, clarifying its connection logic with devices such as busbars, circuit breakers, and disconnectors. Based on the type of operation verb, the module further derives the expected state change path of the target equipment: a tripping operation corresponds to a continuous change path of the opening and closing angle between the moving and stationary contacts, covering the full stroke angle change from closed to open; a voltage regulation operation corresponds to a step-by-step gradual change path of the equipment's output voltage level, including the voltage amplitude and duration of each adjustment. These paths clearly indicate the direction of state change, such as the direction of angle increase in a tripping operation, as well as time constraints, such as the requirement that the tripping operation must complete the angle change from 0 degrees to 90 degrees within a specified time. When the system receives multiple device operation commands and there are execution order dependencies, the module establishes a timing constraint relationship chain of state changes through timing analysis. For example, the opening operation of a certain disconnecting switch is executed first, and then the opening operation of the associated circuit breaker is executed. This ensures that the device operation intention topology network constructed subsequently can accurately map the sequence logic of the actual operation.

[0025] In another preferred embodiment of the present invention, the specific process of defining key nodes in the topology network construction module is as follows: Based on the electrical connections of the power grid, equipment parameters, and historical fault propagation records, a power grid fault propagation chain model is constructed. This model visually presents the path of a fault propagating from the originating device to upstream and downstream related devices through a visual graph, including the direction of fault propagation, the scope of impact, and the response modes of related devices. On this basis, the module calculates the positional weight values ​​of each node in the topology. The calculation process comprehensively considers the number of downstream devices affected by the node in the fault propagation chain, the importance level of the devices, and the probability of fault propagation. The weight value increases with the number of downstream devices and is also affected by the importance of the downstream devices; for example, the weight value of nodes associated with main transformer devices is relatively higher.

[0026] Subsequently, the module locates the starting node by traversing the dependencies in the fault propagation chain model. These nodes are at the source of the fault propagation path and have no upstream equipment affecting them; they are defined as source nodes without upstream dependencies. Next, the module filters nodes based on a preset system protection threshold, marking nodes that simultaneously meet the criteria of being at the starting point of the fault propagation chain and having a position weight exceeding the threshold as critical nodes. These nodes are the focus of fault prevention and condition monitoring. When the power grid topology changes due to the addition, removal, or adjustment of connections of equipment, the module automatically triggers an update mechanism. Based on the changed power grid structure, it recalculates the position weights, dependencies, and starting attributes of each node, synchronously adjusting the set of critical nodes to ensure that the critical nodes always accurately reflect the current fault risk distribution of the power grid, providing a reliable basis for subsequent task selection and computing power allocation.

[0027] In another preferred embodiment of the present invention, the filtering process of the core task set in the task filtering module is as follows: In the device operation intention topology network, the direct dependent nodes of the key nodes are retrieved layer by layer. These dependent nodes are the preceding state nodes that must be associated with to verify the state of the key nodes. These nodes are integrated to form a set of key verification targets, and the state verification objects that need to be monitored are identified.

[0028] Subsequently, the module analyzes the data processing objectives of each task in the perception task set. The device status recognition task primarily processes device image information to output device operating status parameters, while the robot motion control task plans a movement path based on terrain geometry information and outputs position and attitude parameters. By comparing the correlation between the task processing objectives and the set of key verification objectives, tasks that can directly output the physical state parameters of the key verification objective set are selected. These tasks are the foundation for ensuring the status verification of key nodes.

[0029] After initial screening, the module uses a computing power assessment tool to calculate the sum of the minimum computing power guarantees for the selected tasks. This sum reflects the minimum computing power resources required to ensure the status monitoring of all key verification targets. This sum is compared with the real-time available computing power of the edge computing unit. If the sum exceeds the available computing power, it indicates that the current computing power is insufficient to support all the selected tasks. In this case, some tasks are gradually removed in descending order of their key node location weight values ​​until the sum of the minimum computing power guarantees for the remaining tasks is lower than the available computing power. The tasks that are not removed ultimately constitute the core task set, while the removed tasks are converted into background monitoring tasks, and a maximum computing power quota is set for them to limit their consumption of computing power resources.

[0030] In another preferred embodiment of the present invention, the process of calculating the minimum computing power guarantee value for each task in the core task set in the computing power allocation module is as follows: The power grid physical equipment corresponding to the core task set is marked as associated equipment. The basic computing power requirement of the task is calculated based on the voltage level coefficient and historical failure rate of the associated equipment. The basic computing power requirement of the task is relatively higher for equipment with higher voltage levels, and the basic computing power requirement of the task is also increased accordingly for equipment with high historical failure rates, so as to ensure the computing power support for monitoring high-risk equipment.

[0031] Simultaneously, the module analyzes the rate of change of surface curvature in the terrain geometry information. When this rate exceeds a safety threshold, it indicates a complex terrain environment, which increases the data processing difficulty of the robot's motion control task, thus generating a terrain interference component. It also extracts the power frequency noise energy density from the electromagnetic field strength information and generates an electromagnetic interference component based on the distribution characteristics of the energy density. The higher the noise energy density, the larger the electromagnetic interference component, reflecting an increase in the complexity of the equipment's image information processing. The module fuses the terrain interference component and the electromagnetic interference component according to preset rules to form an environmental interference coefficient, comprehensively characterizing the impact of environmental factors on the task's computing power requirements.

[0032] Finally, the basic computing power requirement of the task is multiplied by the environmental interference coefficient. The result is the minimum computing power guarantee value for each task in the core task set. This value takes into account both the importance of monitoring the equipment itself and the impact of environmental interference on computing power requirements, providing an accurate basis for subsequent computing power allocation.

[0033] In a preferred embodiment, the generation process of the environmental interference intensity coefficient in the computing power allocation module is as follows: First, the module analyzes the correlation between the rate of change of surface curvature and the terrain interference component. They exhibit a linear proportional relationship, meaning that the higher the rate of change of surface curvature, the greater the terrain interference component, reflecting the impact of complex terrain on data processing. Simultaneously, the module establishes the relationship between power frequency noise energy density and electromagnetic interference components. Both exhibit an exponential growth characteristic, meaning that as the power frequency noise energy density increases, the growth rate of the electromagnetic interference component gradually increases, demonstrating the significant impact of strong electromagnetic environments on signal analysis. Then, based on the equipment distribution and inspection requirements of different areas of the substation, the module uses preset regional sensitivity weights to fuse the terrain interference and electromagnetic interference components. The final fused result is the environmental interference intensity coefficient, comprehensively reflecting the combined impact of terrain and electromagnetic factors on computing power requirements.

[0034] In another preferred embodiment of the present invention, the specific process of triggering the degradation of non-core task computing functions in the computing power allocation module is as follows: When the actual computing power consumed by the core task set is lower than the minimum guaranteed computing power, the module first detects the terrain conditions of the robot's mobile control task. By analyzing the surface undulation features in the terrain geometry, it determines whether the standard deviation of the current terrain undulation is within the safe threshold range, and simultaneously calculates the obstacle distribution density on the inspection path. If the detection results show that the terrain undulation is gentle and obstacles are sparse, it indicates that the robot's mobile environment is stable. At this time, the real-time 3D point cloud reconstruction function is automatically turned off, and the historical path tracking mode is activated instead to reduce the computing power consumption of the mobile control task.

[0035] Meanwhile, the module categorizes and screens image acquisition tasks, verifying whether the devices associated with each task belong to the key verification target set. For image acquisition tasks whose associated devices are not in the key verification target set, the original continuous video analysis mode is adjusted to a keyframe sampling analysis mode, while extending the sampling interval to reduce computing power consumption by decreasing data processing volume. During the degradation of non-core task functions, the system continuously caches the raw data collected by the sensor array. When the total remaining computing power of the edge computing unit exceeds the preset recovery threshold for a continuous period of time, the post-completion computing mechanism is automatically activated to fully process the cached raw data, ensuring that the monitoring data of non-core tasks is not lost, thus balancing computing power optimization and data integrity.

[0036] In another preferred embodiment of the present invention, the process of generating the equipment state anomaly coefficient in the fault prediction module is as follows: First, obtain the real-time status parameters of the equipment output from the core task set, including equipment operating temperature, mechanical displacement, and electrical parameters. Extract the expected status standard values ​​for the corresponding equipment at different operating stages from the equipment operation intention topology network. These standard values ​​are determined based on equipment design parameters, operating procedures, and historical normal status data. By comparing the real-time status parameters with the expected status standard values, calculate the absolute deviation between the two, record the deviation results for each inspection cycle, and construct a continuous time series of deviations.

[0037] The module employs a sliding window algorithm to analyze the time series of deviations, identifying whether a monotonically increasing trend exists in the series, i.e., a pattern of continuous expansion of the deviation over time. When the cumulative increase of this trend exceeds the allowable tolerance range of the equipment design, the anomaly coefficient calculation process is initiated. The anomaly coefficient calculation comprehensively considers the duration of the trend and the rate of increase of the deviation, exhibiting a non-linear growth characteristic. That is, the longer the trend lasts and the faster the rate of increase, the larger the value of the anomaly coefficient, thereby quantifying the severity of the equipment's deviation from the normal range.

[0038] In a preferred embodiment, the process of generating the maintenance operation sequence in the fault prediction module is as follows: When the abnormal device status coefficient exceeds the preset operation threshold, the module locates the direct upstream device of the device from the device operation intention topology, queries the position weight value of the upstream device in the topology, and compares it with the preset ratio of the system protection threshold.

[0039] If the location weight value of an upstream device exceeds the preset proportion of the system protection threshold, and the growth rate of the anomaly coefficient is higher than the set transmission threshold, it indicates that there is a risk of the fault propagating to upstream devices. At this time, device isolation instructions and backup device switching instructions are generated to prevent the fault from spreading and affecting the operation of related devices. If the upstream device of the device corresponding to the anomaly coefficient is redundantly configured, that is, there is a backup device with the same function, the module calculates the maximum delayed maintenance time window based on the health score of each backup device and the current power grid load. This determines the longest time that maintenance operations can be delayed without affecting the stable operation of the power grid, providing a flexible time reference for operation and maintenance scheduling.

[0040] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent inspection and fault prediction system based on computing power services, characterized in that, include: A sensor array is used to collect environmental data of the substation, including equipment image information, terrain geometry information, and electromagnetic field strength information. The instruction parsing module is used to parse the equipment operation instructions issued by the power grid dispatching system and determine the target operation equipment and its expected state change path. The topology construction module is used to build a device operation intent topology network. The nodes of the topology network represent the physical states that the target operation device needs to verify, and the connecting lines between nodes represent the dependencies of state verification. Nodes located at the beginning of a power grid fault propagation chain and whose location weight exceeds the protection threshold are marked as critical nodes. In the topology construction module, the specific process of defining key nodes is as follows: A power grid fault propagation chain model is constructed, which represents the path relationship of fault propagation from the initiating device to the associated devices. The weight value of the topology node is calculated, and the weight value is proportional to the number of downstream devices affected by the node in the fault propagation chain. Locate the starting node of the fault propagation chain. The starting node is defined as the source node with no upstream dependencies. Nodes that simultaneously meet the criteria of being located at the starting end and having a position weight value exceeding the system protection threshold are marked as critical nodes. When the power grid topology changes, the set of critical nodes is updated again. The process of obtaining position weights includes: Based on the electrical connections of the power grid, equipment parameters, and historical fault propagation records, a power grid fault propagation chain model is constructed. The path relationship of the fault propagation from the starting equipment to the upstream and downstream related equipment is presented intuitively through a visual map, including the direction of fault propagation, the scope of impact, and the response mode of related equipment. Based on this, the position weight value of each node in the topology is calculated. The calculation process takes into account the number of downstream devices affected by the node in the fault propagation chain, the importance level of the devices, and the probability of fault propagation. The weight value increases with the increase of the number of downstream devices. The task filtering module is used to generate a set of perception tasks based on the type of environmental data, including equipment status recognition tasks and robot movement control tasks. It filters out tasks in the perception task set that directly output the physical state parameters of key nodes and marks them as the core task set. The computing power allocation module is used to calculate the minimum computing power guarantee value for each task in the core task set based on the location weight of key nodes and the environmental interference coefficient. When the actual computing power occupied by the core task set is lower than the sum of the minimum computing power guarantee values, the non-core task computing function is degraded. The process of obtaining the environmental interference coefficient includes: The rate of change of surface curvature in the topographic geometry information is analyzed. When the rate exceeds the safety threshold, it indicates that the topographic environment is complex and a topographic interference component is generated. The power frequency noise energy density in the electromagnetic field strength information is extracted. An electromagnetic interference component is generated based on the distribution characteristics of the energy density. The topographic interference component and the electromagnetic interference component are fused according to a preset rule to form an environmental interference coefficient. The fault prediction module is used to input the equipment status parameters output by the core task set into the fault prediction network to generate equipment status anomaly coefficients and maintenance operation sequences.

2. The intelligent inspection and fault prediction system based on computing power services according to claim 1, characterized in that, The process of parsing equipment operation instructions issued by the power grid dispatching system in the instruction parsing module is as follows: It receives standardized operation command streams issued by the power grid dispatching system. The command streams conform to the power system operation protocol specifications. It parses the command text structure and extracts the pairing relationship between operation verbs and equipment identifiers. The operation verbs include tripping operation, closing operation, and voltage regulation operation types. The system queries the power grid topology database based on the device identifier to obtain the physical installation location of the device and its upstream and downstream relationships in the electrical connection diagram; it derives the expected state change path based on the operation verb type, such as the opening and closing angle change path corresponding to the tripping operation and the voltage level change path corresponding to the voltage regulation operation, both of which include the state change direction and time constraints. When multiple device operation commands have execution order dependencies, a time-series constraint relationship chain for state changes is established to ensure that the dependency relationship of the device operation intention topology accurately reflects the operation sequence logic.

3. The intelligent inspection and fault prediction system based on computing power services according to claim 1, characterized in that, In the task filtering module, the filtering process for the core task set is as follows: In the device operation intention topology network, identify the direct dependent nodes of key nodes to form a set of key verification targets; analyze the data processing targets of each task in the perception task set, and select the tasks that can output the physical state parameters of the key verification target set. Calculate the total minimum computing power guarantee value of the selected tasks; The total is compared with the real-time available computing power of the edge computing unit. When the total exceeds the available computing power, tasks are removed in descending order of location weight value until the computing power constraint is met. The tasks that are not removed are set as the core task set, and the removed tasks are converted into background monitoring tasks and set with a maximum computing power quota.

4. The intelligent inspection and fault prediction system based on computing power services according to claim 1, characterized in that, In the computing power allocation module, the process of calculating the minimum computing power guarantee value for each task in the core task set is as follows: Mark the power grid physical equipment corresponding to the core task set as associated equipment, and calculate the basic computing power requirement of the task based on the voltage level coefficient and historical failure rate of the associated equipment. The terrain geometry information is analyzed to determine the rate of change of surface curvature. When the rate of change exceeds a safety threshold, a terrain interference component is generated. The power frequency noise energy density is extracted from the electromagnetic field strength information, and an electromagnetic interference component is generated based on the energy density distribution. The terrain interference component and the electromagnetic interference component are then fused to form an environmental interference intensity coefficient. Multiply the basic computing power requirement of the task by the environmental interference intensity coefficient, and output the product as the minimum computing power guarantee value.

5. The intelligent inspection and fault prediction system based on computing power services according to claim 4, characterized in that, In the computing power allocation module, the process for generating the environmental interference intensity coefficient is as follows: A linear proportional function is established between the rate of change of surface curvature and the terrain interference component, and an exponential growth function is established between the power frequency noise energy density and the electromagnetic interference component. The two types of interference components are fused according to the preset regional sensitivity weights, and the fusion result is the environmental interference intensity coefficient.

6. The intelligent inspection and fault prediction system based on computing power services according to claim 1, characterized in that, In the computing power allocation module, the specific process of triggering the degradation of non-core task computing functions is as follows: When the actual computing power occupied by the core task set is lower than the total guaranteed value, the terrain conditions of the robot's mobile control task are detected. If the standard deviation of terrain undulation is lower than the threshold and the obstacle density is sparse, real-time 3D point cloud reconstruction is turned off and historical path tracking is activated. The system detects whether the image acquisition task associated with the detection equipment is outside the set of key verification targets. If it is outside the set, the continuous video analysis is converted into keyframe sampling analysis and the sampling interval is extended. During the degradation process, the original sensor data is continuously cached. When the total remaining computing power of the edge computing unit is continuously higher than the recovery threshold, the cached data is supplemented by calculation.

7. The intelligent inspection and fault prediction system based on computing power services according to claim 1, characterized in that, In the fault prediction module, the process of generating the equipment status anomaly coefficient is as follows: Obtain the real-time status parameters of the core task set monitoring equipment, extract the expected status standard values ​​of the corresponding equipment from the equipment operation intention topology network, calculate the absolute deviation between the real-time parameters and the standard values, and construct the deviation time series of continuous inspection cycles. A sliding window algorithm is used to identify monotonically increasing trends in time series. When the cumulative increase of the increasing trend exceeds the equipment design tolerance range, the anomaly coefficient calculation is initiated. The anomaly coefficient of non-linear growth is calculated based on the trend duration and the rate of increase.

8. The intelligent inspection and fault prediction system based on computing power services according to claim 7, characterized in that, In the fault prediction module, the process of generating the maintenance operation sequence is as follows: When the abnormal coefficient exceeds the operation threshold, locate the direct upstream device of the device corresponding to the abnormal coefficient in the topology network, and determine whether the position weight value of the upstream device exceeds the preset ratio of the system protection threshold. If the weight value exceeds the preset proportion of the system protection threshold and the abnormal coefficient growth rate is higher than the transmission threshold, a device isolation and backup switching instruction is generated; if the upstream device is redundantly configured, the maximum delayed maintenance time window is calculated based on the health score.

Images