Substation autonomous inspection method based on multi-source information fusion, mobile robot and system

Abstract

The present application relates to the technical field of substation automatic inspection, and particularly relates to a substation autonomous inspection method based on multi-source information fusion, a mobile robot and a system. The present application combines inspection subareas into an inspection block, selects a target block of the mobile robot according to the inspection demand degree of the inspection subareas in the inspection block and the distance between the real-time position of the mobile robot and the inspection block, selects a current target device of the mobile robot and goes to inspection, generates an inspection path based on the historical inspection associated features and environmental state features of the current target device, and executes a detection task on the inspection items along the path, reselects a new current target device of the mobile robot after completing the inspection of the current target device, until the remaining un-inspected target devices in the target block meet a preset exit condition, and selects a new target block of the mobile robot. The present application solves the lag of hidden danger investigation and improves the inspection efficiency through macro selection, medium focus on devices and micro customized path.

Description

Technical Field

[0001] This invention relates to the field of automatic substation inspection technology, specifically to a substation autonomous inspection method, mobile robot, and system based on multi-source information fusion. Background Technology

[0002] By using mobile robots to replace manual inspection tasks, substations can effectively reduce the labor intensity and operational safety risks of maintenance personnel, achieve automated monitoring of equipment status around the clock, improve the timeliness of defect detection and the level of intelligent operation and maintenance, and thus provide a reliable guarantee for the safe and stable operation of the power grid.

[0003] The prior art, patent document CN107560631A, discloses a path planning method, device, and inspection robot. Specifically, it initializes predetermined stopping points and a fixed task list, reads the shortest distance matrix, and then uses an improved ant colony algorithm to calculate the spatial distance path between the task points in the preset task list. Finally, it refines the route using physical constraints such as one-way streets to obtain the task point inspection sequence. However, the existing solution mainly solves the shortest spatial path based on a static preset task list. Its computational premise is disconnected from multi-source information such as equipment operating status and surrounding environment, making it impossible to perceive the actual health degradation trend of the equipment during inspection task generation. Furthermore, because such algorithms often use traversing preset nodes and finding the shortest path as the sole optimization objective, they often adopt an average allocation strategy for high-risk and low-risk equipment, causing the robot to fail to prioritize high-risk equipment, resulting in a serious waste of inspection resources and significant lag and limitations in the investigation of potential hazards in high-risk equipment. Summary of the Invention

[0004] To address the technical problems of rigid resource allocation in inspections leading to severe delays in hazard identification and low overall inspection efficiency, this invention aims to provide a substation autonomous inspection method, mobile robot, and system based on multi-source information fusion. The specific technical solution adopted is as follows: In a first aspect, one embodiment of the present invention provides a method for autonomous inspection of substations using multi-source information fusion, the method comprising: The system acquires multi-source inspection data of each piece of equipment to be inspected in the substation in real time, as well as the real-time location of each mobile robot, and divides the inspection area of ​​the substation into different inspection zones. Based on multi-source inspection data, update the risk pheromones of each device to be inspected; based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval, obtain the inspection demand of each inspection zone, and merge the inspection zones into different inspection blocks. Based on the inspection demand of each inspection zone and the distance between the mobile robot's real-time position and the inspection zone, the target zone of the mobile robot is selected, and the mobile robot is controlled to move to the target zone. Based on the risk pheromones and spatial location characteristics of the uninspected devices in the target block, the mobile robot selects the current target device and proceeds to inspect it; after completing the inspection of the current target device, the mobile robot selects a new current target device, and the inspection of the target block ends when the remaining uninspected devices in the target block meet the preset exit conditions, and a new target block is selected for the mobile robot. The inspection process for the current target equipment includes: generating an inspection path based on the historical inspection association features and environmental status features of the current target equipment, and performing detection tasks on the inspection items along the inspection path to complete the inspection of the current target equipment.

[0005] Furthermore, updating the risk pheromones of each device under inspection based on multi-source inspection data includes: Based on the life cycle stage of each device to be inspected, an initial risk pheromone is set for the corresponding device; The multi-source inspection data includes: different types of abnormal events of the equipment under inspection; For each device under inspection, when the multi-source inspection data of the device under inspection at the current moment contains an abnormal event, the time interval between the abnormal event of the device under inspection and the most recent abnormal event of the same type is obtained and recorded as the abnormal reproduction time; the abnormal reproduction time is negatively correlated and normalized, and the sum of the processing result and the constant 1 is used as the pheromone adjustment coefficient. When the multi-source inspection data of the device under inspection does not contain any abnormal events at the current moment and the device under inspection has completed the inspection, the number of inspections in which no abnormalities were found between the current moment and the most recent abnormal event is obtained, and the number of inspections is negatively correlated and normalized to obtain the pheromone adjustment coefficient. The risk pheromone of the device under inspection is multiplied by the pheromone adjustment coefficient to update the risk pheromone and obtain the updated risk pheromone of the device under inspection.

[0006] Furthermore, the step of obtaining the inspection demand level of each inspection zone and merging the inspection zones into different inspection blocks includes: For each inspection zone, the maximum value of the risk pheromone of all devices to be inspected within the inspection zone is extracted; the multi-source inspection data includes the historical inspection completion time of each device to be inspected; the time difference between the current time and the most recent inspection completion time of the inspection zone is taken as the historical inspection time interval; the product of the historical inspection time interval and the maximum value is calculated as the inspection demand degree of the inspection zone. The demand level of each inspection zone within the inspection area is determined based on the demand for inspection. Region growth is performed by taking any inspection zone within the inspection area as a growth point. Inspection zones with the same demand level as the growth point within the preset neighborhood of the growth point are taken as new growth points for region growth. Region growth stops when there are no inspection zones with the same demand level as the growth point within the preset neighborhood of all new growth points, resulting in different inspection blocks.

[0007] Further, selecting the target block for the mobile robot includes: The demand level includes a high demand level; the percentage of inspection zones at the high demand level in each inspection block is counted; the average value of the demand level of all inspection zones in each inspection block is calculated, and the sum of the percentage of the number and a constant 1 is multiplied by the average value to obtain the block urgency of the corresponding inspection block. Obtain the distance from the real-time position of the mobile robot to the geometric center of each inspection block, and use the ratio of the block urgency to the distance as the block selection weight for the corresponding inspection block; Based on the block selection weights of all inspection blocks, the block selection weights of each inspection block are normalized to obtain the block selection probability; based on the block selection probability, the roulette wheel algorithm is used to select the target block of the mobile robot from all inspection blocks.

[0008] Furthermore, the selection of the current target device for the mobile robot includes: Record the uninspected devices within the target block of the mobile robot as candidate devices at the current moment; obtain the shortest distance from each candidate device within the target block to the boundary of the target block, and the straight-line distance from the real-time position of the mobile robot to each candidate device at the current moment. Based on the risk pheromone of each candidate device, the shortest distance, and the straight-line distance, the device candidate weight of the corresponding candidate device is obtained; the risk pheromone and the shortest distance are both positively correlated with the device candidate weight, and the straight-line distance is negatively correlated with the device candidate weight; Based on the device candidate weights of all candidate devices within the target block, the device candidate weights of each candidate device are normalized to obtain the device selection probability; based on the device selection probability, the roulette wheel algorithm is used to select the current target device from all candidate devices.

[0009] Further, the step of generating an inspection path and performing a detection task on the inspection items along the inspection path includes: A pre-constructed association graph of inspection items for different device types is constructed, including different inspection item nodes connected by directed edges, wherein the directed edges are configured with initial path pheromones; The source node of the inspection item association graph of the current target device type is used as the initial current inspection node, and the source node is used as the starting point of the target path; the inspection item node pointed to by the directed edge starting from the current inspection node is used as the candidate node of the current inspection node. Multi-source inspection data includes environmental data; environmental state sensitivity is determined based on the environmental data at the current time of the current target device's location; the node jump weight of the candidate node is obtained and normalized according to the path pheromone of the directed edges pointing to each candidate node of the current inspection node, the number of times the candidate node has been inspected in this inspection, and the environmental state sensitivity, so as to obtain the node jump probability. Based on the node jump weight, an inspection path is generated in the inspection item association graph corresponding to the current target device; The inspection task is executed sequentially for each inspection item node along the inspection path. For each inspection item node on the inspection path, if the inspection result of the inspection item node is abnormal, the number of times the inspection item node has not been selected into the inspection path since the last abnormality was detected is obtained. The path pheromone of the directed edge pointing to the inspection item node is adjusted using the number of times, and the adjusted path pheromone of the directed edge is obtained to update the inspection item association graph corresponding to the current target device. The updated inspection item association graph is then used as the inspection item association graph of the current target device in the next inspection.

[0010] Furthermore, the generation of the inspection path in the inspection item association graph corresponding to the current target device based on node jump weight includes: Select candidate nodes from the current check node whose node jump probability is greater than or equal to the preset jump lower limit and record them as valid jump nodes; If there is a valid jump node at the current inspection node, a check item node is selected from the valid jump nodes based on the node jump probability and added to the target path for updating, and the selected check item node is used as the new current inspection node. If there are no candidate nodes or valid jump nodes at the current inspection node, then backtrack to the previous inspection item node in the target path and use that inspection item node as the new current inspection node. The update of the target path stops when the path is traced back to the source point and there are no candidate nodes or valid jump nodes at the source point. The updated target path is then used as the inspection path.

[0011] Furthermore, the preset exit conditions include: the proportion of the number of candidate devices in the target block is less than a preset lower limit for inspection, and the risk pheromones of all candidate devices are less than a preset lower limit for pheromones.

[0012] A mobile robot includes a memory and a processor, wherein the memory stores instructions; the processor invokes the instructions in the memory to cause the mobile robot to perform the steps of the substation autonomous inspection method based on multi-source information fusion as described above.

[0013] Another embodiment of the present invention provides a substation autonomous inspection system based on multi-source information fusion, the system comprising: The data acquisition module is used to acquire multi-source inspection data of each piece of equipment to be inspected in the substation and the real-time position of each mobile robot in real time, and to divide the inspection area of ​​the substation into different inspection zones. The inspection block generation module is used to update the risk pheromones of each device to be inspected based on multi-source inspection data; obtain the inspection demand degree of each inspection zone based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval; and merge the inspection zones into different inspection blocks. The target block selection module is used to select the target block of the mobile robot based on the inspection demand of the inspection area in each inspection block and the distance between the real-time position of the mobile robot and the inspection block, and control the mobile robot to move to the target block. The target device selection module is used to select the current target device of the mobile robot and proceed to inspect it based on the risk pheromone and spatial location characteristics of the uninspected devices in the target block; after completing the inspection of the current target device, the mobile robot selects a new current target device, and the inspection of the target block ends when the remaining uninspected devices in the target block meet the preset exit conditions, and a new target block is selected for the mobile robot. The inspection process for the current target equipment includes: generating an inspection path based on the historical inspection association features and environmental status features of the current target equipment, and performing detection tasks on the inspection items along the inspection path to complete the inspection of the current target equipment.

[0014] The present invention has the following beneficial effects: In this embodiment of the invention, the inspection demand of each inspection zone is determined based on the risk pheromones of the equipment to be inspected within the inspection zone and the historical inspection time interval, so as to accurately measure the urgency of inspection in each area, and inspection zones with similar demands are merged into inspection blocks; the target block is selected by comprehensively considering the inspection demand and the distance between the real-time position of the mobile robot and the inspection block, so as to achieve a dynamic balance between the urgency of risk investigation and the execution cost of movement, and significantly improve the overall inspection efficiency from a global perspective; based on the risk pheromones and spatial location characteristics of the uninspected equipment to be inspected, the current target equipment is adaptively selected for priority investigation; at the same time, pre- By setting exit conditions to promptly end the inspection of the current target block, the robot is guided to accurately focus on high-risk hazards within the block, while preventing excessive time spent on low-risk equipment. This significantly accelerates the turnover of inspection resources and completely solves the problem of severe lag in the investigation of high-risk hazards. Based on the historical inspection correlation characteristics and environmental status characteristics of the current target equipment, a targeted inspection path is dynamically generated, and the inspection items are executed along the inspection path. This ensures that high-risk parts that are susceptible to environmental changes and historical hazards are thoroughly inspected, completely breaking through the efficiency bottleneck caused by rigid inspection of single equipment and greatly improving the timeliness of hazard discovery and overall inspection efficiency. Attached Figure Description

[0015] Figure 1 The flowchart illustrates the steps of a substation autonomous inspection method based on multi-source information fusion, as provided in one embodiment of the present invention. Figure 2 This is a flowchart illustrating a method for obtaining inspection blocks according to an embodiment of the present invention. Figure 3 This is a system structure diagram of a substation autonomous inspection system based on multi-source information fusion, provided as an embodiment of the present invention. Detailed Implementation

[0016] The following detailed description, in conjunction with the accompanying drawings, illustrates the specific scheme of the multi-source information fusion substation autonomous inspection method, mobile robot, and system provided by this invention.

[0017] Example 1: This invention proposes a substation autonomous inspection method based on multi-source information fusion. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a substation autonomous inspection method based on multi-source information fusion, according to an embodiment of the present invention. The method includes: Step S1: Real-time acquisition of multi-source inspection data of each device to be inspected in the substation and the real-time location of each mobile robot, and division of the substation inspection area into different inspection zones.

[0018] Specifically, this solution adopts a dual-mode data update mechanism that combines periodic synchronization with event-triggered synchronization: on the one hand, multi-source inspection data is periodically collected from various data sources every 15 minutes to maintain the regular updates of the background database; on the other hand, when the substation dispatch automation system or equipment online monitoring device generates a new equipment abnormality alarm event, multi-source inspection data is collected immediately to ensure the real-time response priority of high-risk events and avoid the alarm information within the periodic update interval being delayed in processing.

[0019] The multi-source inspection data originates from three types of data: First, real-time operating parameters and abnormal alarm signals of the equipment are obtained through the substation dispatch automation system and online equipment monitoring devices; second, historical inspection records, including timestamps of each inspection, inspection item execution records, and defect descriptions, are extracted from the substation production management system and robot backend server; third, environmental sensor data such as ambient temperature, humidity, rainfall, water depth, and specific gas concentrations are obtained through the station's auxiliary control system. These three data sources, after being aligned with equipment codes and timestamps, together constitute the multi-source inspection data of the equipment under inspection at the current moment.

[0020] The navigation and positioning module onboard the mobile robot acquires the spatial coordinates of each robot in real time. Simultaneously, the substation inspection area is discretized into a square grid with a preset side length (preferably 20 meters). Based on the latest actual spatial coordinates of each device under inspection, the entire substation inspection area is divided into several inspection zones, providing a unified spatial data foundation for subsequent overall station-wide demand assessment and robot area scheduling. The preset side length is 20 meters, but implementers can adjust it according to specific circumstances.

[0021] Step S2: Update the risk pheromones of each device to be inspected based on the multi-source inspection data; obtain the inspection demand degree of each inspection zone based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval, and merge the inspection zones into different inspection blocks.

[0022] Traditional substation inspections often employ fixed cycles and routes, failing to adapt to the dynamic evolution of equipment health status. To break this rigid resource allocation model, this step first dynamically updates the risk pheromones of the equipment to be inspected using multi-source inspection data to quantify the real-time potential hazard level. Then, the risk pheromones are integrated with historical inspection time intervals to determine the inspection demand level of each inspection zone, accurately measuring the urgency of inspection in each area. Inspection zones with similar demands are merged into consecutive inspection blocks. This allows for a global perspective on the true risk distribution of the substation, guiding mobile robots to adaptively concentrate inspection resources on high-risk and long-neglected areas. This effectively solves the technical problems of severely delayed hazard identification and low overall inspection efficiency caused by rigid resource allocation in traditional inspections.

[0023] Step S3: Based on the inspection demand of each inspection zone and the distance between the real-time position of the mobile robot and the inspection zone, select the target zone for the mobile robot and control the mobile robot to move to the target zone.

[0024] To overcome the inefficiency and time waste caused by traditional fixed inspection routes, this system dynamically selects target blocks and guides the robot to them by combining the overall inspection demand of each inspection area with the spatial distance from the real-time location of the mobile robot to that block. This ensures that high-risk and urgently needed areas are prioritized for inspection, while effectively avoiding unnecessary long-distance cross-area travel by the mobile robot through the introduction of spatial distance constraints. This scheduling mechanism achieves a dynamic balance between the urgency of risk assessment and the cost of movement, significantly reducing the wasted movement losses of the robot and thus significantly improving the overall inspection efficiency of the substation's limited resources.

[0025] Step S4: Based on the risk pheromones and spatial location characteristics of the uninspected devices in the target block, select the current target device of the mobile robot and proceed to inspect it; after completing the inspection of the current target device, select a new current target device for the mobile robot, until the remaining uninspected devices in the target block meet the preset exit conditions, then end the inspection of the target block and select a new target block for the mobile robot; the inspection process for the current target device includes: generating an inspection path based on the historical inspection association characteristics and environmental state characteristics of the current target device, and performing inspection tasks on the inspection items along the inspection path to complete the inspection of the current target device.

[0026] After a mobile robot enters a target area, the traditional fixed-sequence, step-by-step inspection method cannot distinguish the risk differences between devices. Therefore, based on the risk pheromones and spatial location characteristics of uninspected devices, the robot adaptively selects the current target device for priority inspection. This process continues after the inspection is completed until the remaining uninspected devices meet preset exit conditions, at which point the current block inspection ends and the robot moves to a new block. This achieves dynamic focusing of the device-level inspection path at the meso-level, ensuring timely and accurate inspection of high-risk devices, avoiding excessive time spent by the robot on low-risk devices, significantly accelerating the flow efficiency of inspection resources across the entire site, and thus solving the technical problems of severely delayed hazard identification and low overall efficiency.

[0027] In the process of single-point inspection of specific equipment, traditional fixed checklists often lead to insufficient investigation of high-risk areas and redundant testing of low-risk areas. To address this, this step delves into the micro-level, dynamically generating targeted and personalized inspection paths based on the historical inspection correlation characteristics and current environmental conditions of the target equipment. This guides the inspection tasks, significantly reducing the time spent on low-value inspection items and ensuring that high-risk areas susceptible to environmental changes and historical conditions are thoroughly investigated. This completely breaks through the efficiency bottleneck caused by rigid single-equipment inspections, greatly improving the timeliness of hazard discovery and overall inspection efficiency.

[0028] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining risk pheromones includes: setting an initial risk pheromone for each device under inspection according to its lifecycle stage; multi-source inspection data including different types of abnormal events of the devices under inspection; for each device under inspection, when the multi-source inspection data of the device under inspection at the current moment contains an abnormal event, obtaining the time interval between the abnormal event of the device under inspection and the most recent abnormal event of the same type, and recording it as the abnormal reproduction time; performing negative correlation and normalization processing on the abnormal reproduction time, and normalizing the negative correlation. The sum of the processed result after normalization and the constant 1 is used as the pheromone adjustment coefficient. When the multi-source inspection data of the device under inspection at the current moment does not contain any abnormal events and the device under inspection has completed its inspection, the number of inspections in which no abnormalities were found between the current moment and the most recent abnormal event is obtained. These inspection numbers are then negatively correlated and normalized, and the result of this negative correlation and normalization is used as the pheromone adjustment coefficient. The risk pheromone of the device under inspection is multiplied by the pheromone adjustment coefficient to update the risk pheromone, resulting in the updated risk pheromone of the device under inspection. It is important to note that for devices under inspection that have never experienced the same type of abnormal event, the initial commissioning time of the device under inspection is used instead of the time of the most recent abnormal event. If the pheromone adjustment coefficient is less than 0.01, it is set to 0.01 to prevent the device from being completely forgotten due to long-term normal operation, thus failing to attract attention again when a sudden abnormality occurs.

[0029] It should be noted that, due to differences in the basic failure rate of equipment at different stages of its life cycle, the time difference between the current moment and the initial commissioning time of the equipment under inspection is used as the running time to provide an accurate risk assessment benchmark. In this embodiment, the equipment life cycle stages include: commissioning break-in period, stable operation period, and aging and degradation period. When the running time is less than 1 year, the equipment is in the commissioning break-in period, with a high initial rate of potential failures, and the initial risk pheromone is 1.5. When the running time is greater than or equal to 1 year but less than 15 years, the equipment is in the stable operation period, with the lowest rate of accidental failures, and the initial risk pheromone is 1. When the running time is greater than or equal to 15 years, the equipment is in the aging and degradation period, with the highest risk of component aging and wear, and the initial risk pheromone is 2. The above values ​​can be set according to specific circumstances. The higher the risk pheromone, the higher the potential defect risk of the equipment under inspection and the greater the required level of inspection attention.

[0030] During equipment operation, its health status dynamically evolves with various events. Therefore, it is necessary to dynamically update risk pheromones based on abnormal events in multi-source inspection data. Types of abnormal events include: transformer oil chromatography anomalies, circuit breaker sulfur hexafluoride pressure low anomalies, surge arrester resistive current anomalies, and protection device self-test anomalies. When multi-source inspection data contains abnormal events, it indicates a deterioration in equipment health. The recurrence frequency of the same type of abnormal event directly reflects the severity of the defect; the shorter the anomaly recurrence time, the more frequently the defect occurs and the more significant the deterioration trend. Therefore, a negative correlation and normalization process is applied to the anomaly recurrence time. Specifically, the maximum value of the anomaly recurrence time in historical statistics for the same type of equipment is used as the upper bound of the normalization reference. The ratio of the anomaly recurrence time to the upper bound of the normalization reference is calculated. A negative correlation and normalization process is achieved by subtracting a constant 1 from this ratio. The sum of this negatively correlated and normalized result and a constant 1 is used as the pheromone adjustment coefficient, ensuring that this coefficient is greater than 1, to positively amplify the risk pheromone. When multi-source inspection data does not contain any abnormal events and the equipment under inspection has completed its inspection, it indicates that the equipment is currently operating stably. To free up robot inspection resources and avoid over-inspection, risk pheromones need to be attenuated. The greater the number of consecutive inspections without abnormalities detected for the equipment under inspection, the more reliable its status, and the lower the required inspection attention. Therefore, the number of inspections is negatively correlated and normalized. Specifically, the maximum number of consecutive normal inspections allowed for similar equipment as specified in the standard inspection criteria is used as the upper limit of normalization. The ratio of the number of inspections to the upper limit of normalization is calculated. By subtracting the constant 1 from this ratio, a negative correlation is achieved and normalization is performed, resulting in a pheromone adjustment coefficient between 0 and 1, which is used to attenuate risk pheromones in the opposite direction. Finally, the product of the risk pheromone of the equipment under inspection and the pheromone adjustment coefficient is used as the updated risk pheromone, thus completing the adaptive dynamic evolution of risk pheromones according to the real-time status of the equipment. It should be noted that if an abnormal event occurs for the first time for the equipment under inspection, the abnormality reproduction time is set to 1 hour. Additionally, it should be noted that if the time for an anomaly to reproduce is greater than its corresponding normalized reference upper bound, the ratio of the anomaly to the normalized reference upper bound will be forcibly set to 1; if the number of inspections is greater than its corresponding normalized reference upper bound, the ratio of the number of inspections to the normalized reference upper bound will be forcibly set to 1.

[0031] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the inspection block is described in [reference needed]. Figure 2 The diagram illustrates a flowchart of a method for obtaining inspection blocks according to an embodiment of the present invention. The method includes: Step S210: For each inspection zone, extract the maximum value of the risk pheromone of all devices to be inspected within the inspection zone; multi-source inspection data includes the historical inspection completion time of each device to be inspected; take the time difference between the current time and the most recent inspection completion time of the inspection zone as the historical inspection time interval; calculate the product of the historical inspection time interval and the maximum value as the inspection demand degree of the inspection zone.

[0032] It should be noted that the maximum value of the risk pheromone reflects the peak degree of equipment defect deterioration within the inspection zone. A higher value indicates that at least one piece of equipment within the inspection zone is in a state of continuous defect worsening, and a failure would directly threaten the safe operation of the substation, thus requiring a higher inspection intervention priority. A longer historical inspection interval means that hidden defects such as insulation aging, mechanical wear, and environmental corrosion have accumulated during this period without timely investigation, and the uncertainty of failure increases with the duration of the delay. In summary, the maximum value of the risk pheromone and the historical inspection interval are both positively correlated with the degree of inspection demand. The higher the degree of inspection demand, the higher the inspection priority allocated to the inspection zone in macro-resource scheduling should be.

[0033] It is important to note that if an inspection zone has never been inspected at the current time, the historical inspection time interval is assigned the preset maximum tolerance time. This preset maximum tolerance time is based on the longest routine inspection cycle specified in the substation operation and maintenance regulations; in this embodiment, it is set to 720 hours. This aims to give the uninspected blind spot the highest time weight to force priority inspection. If the time difference between the current time and the most recent inspection completion time of the inspection zone is less than the time offset compensation constant, the historical inspection time interval is set to the time offset compensation constant. This time offset compensation constant is set based on the basic time consumption of a single complete inspection task; in this embodiment, it is set to 1 hour. This aims to prevent the calculation results of the inspection demand from becoming invalid when the inspection zone has just completed its inspection due to the time approaching zero.

[0034] Step S220: Determine the demand level of each inspection zone within the inspection area based on the demand for inspection.

[0035] In this embodiment of the invention, the method for obtaining the demand level of an inspection zone includes: obtaining the upper quartile and lower quartile of the demand degree of all inspection zones within the inspection area; determining whether the difference between the upper quartile and the lower quartile is greater than a preset minimum difference; if so, recording the upper quartile and the lower quartile as the high demand threshold and the low demand threshold, respectively; if not, calculating the average value of the demand degree of all inspection zones within the inspection area, and recording the sum and difference of the average value and the preset minimum difference as the high demand threshold and the low demand threshold, respectively; setting the demand level of inspection zones with a demand degree greater than the high demand threshold as the high demand level, setting the demand level of inspection zones with a demand degree less than or equal to the high demand threshold and greater than or equal to the low demand threshold as the medium demand level, and setting the demand level of inspection zones with a demand degree less than the low demand threshold as the low demand level.

[0036] Step S230: Use any inspection partition within the inspection area as a growth point to perform region growth. Use inspection partitions within the preset neighborhood of the growth point with the same demand level as the growth point as new growth points to perform region growth. Stop region growth when there are no inspection partitions within the preset neighborhood of all new growth points with the same demand level as the growth point, and obtain different inspection blocks.

[0037] It should be noted that the demand level represents the overall urgency of the inspection of the inspection zone. In order to ensure the consistency of the operation and maintenance goals in the merged work area, the inspection zone with the same demand level as the growth point within the preset neighborhood is selected as the new growth point to continue to spread outward and search, so as to obtain the inspection block used to guide the mobile robot to carry out macro-level resource concentration and continuous operation.

[0038] In this embodiment of the invention, the inspection zone is regarded as a pixel, and the preset neighborhood is set to eight neighborhoods. The implementer can set it according to the specific situation.

[0039] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the target block includes: demand levels including high demand levels; statistically analyzing the proportion of inspection partitions at the high demand level within each inspection block; calculating the average value of the uninspected demand degree of all inspection partitions within each inspection block, multiplying the sum of the proportion of the number and a constant 1 by the average value to obtain the block urgency of the corresponding inspection block; obtaining the distance from the real-time position of the mobile robot to the geometric center of each inspection block, and using the ratio of the block urgency to the distance as the block selection weight of the corresponding inspection block; normalizing the block selection weight of each inspection block based on the block selection weights of all inspection blocks to obtain the block selection probability; and selecting the target block of the mobile robot from all inspection blocks using a roulette wheel algorithm based on the block selection probability.

[0040] It should be noted that the average demand level of the inspection sub-areas within an inspection block reflects the overall defect risk level and urgency of the inspection. To avoid the average value masking the true clustering of local high-risk points, the average value is adjusted using the proportion of inspection sub-areas with high demand levels within the inspection block. This results in a block urgency level that characterizes the degree of urgency for the robot to prioritize a comprehensive inspection of the inspection block; the higher the block urgency, the higher the inspection priority of the block. To prevent increased energy consumption and reduced inspection efficiency due to long-distance cross-area movement of the mobile robot, spatial distance is introduced as a penalty term. Specifically, the ratio of the block urgency to the distance from the real-time position of the mobile robot to the geometric center of each inspection block is used as the block selection weight. The larger the value, the more urgent the inspection task of the inspection block is and the closer the physical distance, making it more valuable for priority. Since the block selection weight is only a relative score, the block selection weight of each inspected block needs to be divided by the sum of the block selection weights of all inspected blocks to normalize the block selection weights and obtain the block selection probability. To prevent scheduling from getting stuck in local deadlock and to retain global exploration capability, a roulette wheel algorithm is used to select the target block for the mobile robot based on the block selection probability. The roulette wheel algorithm is a well-known technique to those skilled in the art and will not be described in detail here.

[0041] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the current target device includes: recording uninspected devices within the target block of the mobile robot as candidate devices at the current moment; obtaining the shortest distance from each candidate device within the target block to the boundary of the target block, and the straight-line distance from the real-time position of the mobile robot to each candidate device at the current moment; obtaining the device candidate weight of each candidate device based on the risk pheromone, shortest distance, and straight-line distance; the risk pheromone and shortest distance are positively correlated with the device candidate weight, and the straight-line distance is negatively correlated with the device candidate weight; normalizing the device candidate weight of each candidate device based on the device candidate weight of all candidate devices within the target block to obtain the device selection probability; and selecting the current target device from all candidate devices using a roulette wheel algorithm based on the device selection probability.

[0042] It should be noted that, to avoid redundant work, any uninspected devices within the target block at the current moment are designated as candidate devices. To balance movement costs and the completeness of block inspection coverage, the shortest distance from the candidate device to the target block boundary, as well as the straight-line distance from the mobile robot's real-time position to the candidate device at the current moment, are obtained. A larger shortest distance indicates that the candidate device is located deeper within the target block; a larger straight-line distance indicates higher energy and time costs for the robot to reach the candidate device. To guide the robot to prioritize inspecting high-risk devices nearby within the block, while preventing the robot from only superficially inspecting the block edges and missing devices deep within the block, device candidate weights are determined based on the aforementioned distance characteristics and the risk pheromones of the candidate devices. Among them, the higher the risk pheromone, the more serious the potential defects of the equipment, and the stronger the demand for priority inspection. Therefore, the risk pheromone is positively correlated with the candidate weight of the equipment. Prioritizing the inspection of equipment deep in the hinterland can prevent the robot from going beyond the block boundary too early and ensure the integrity of the current block coverage. Therefore, the shortest distance is positively correlated with the candidate weight of the equipment. Prioritizing the inspection of nearby equipment can reduce unnecessary round trips and reduce mobile energy consumption. Therefore, the straight-line distance is negatively correlated with the candidate weight of the equipment.

[0043] In this embodiment of the invention, the product of the reciprocal of the straight-line distance from the real-time position of the mobile robot to each candidate device at the current moment, the risk pheromone of the candidate device, and the shortest distance from the candidate device to the boundary of the target block within the mobile robot's target block is used as the device candidate weight for the corresponding candidate device. A larger device candidate weight reflects that the candidate device has a higher overall value for priority inspection at the current moment. It should be noted that, to prevent the risk pheromone of candidate devices located on the block boundary from being zeroed out during the multiplication operation due to the shortest distance approaching zero, if the shortest distance is less than a preset lower boundary distance limit, the shortest distance is assigned to the preset lower boundary distance limit. Specifically, it is determined whether the number of candidate devices within the target block is greater than 1. If so, the preset lower boundary distance limit is set to one-tenth of the average Euclidean distance between all pairwise candidate devices within the target block; otherwise, the preset lower boundary distance limit is assigned to 0.1 meters.

[0044] Since the device candidate weights are only relative scores, to prevent path planning from getting stuck in local deadlocks and to retain global exploration capabilities, the sum of the device candidate weights of all candidate devices within the target block of the mobile robot is calculated. The device candidate weights are then normalized by calculating the ratio of each candidate device's weight to the sum, yielding the device selection probability for each candidate device. Based on these device selection probabilities, a roulette wheel algorithm is used to select the current target device from all candidate devices, causing the robot to move towards higher-value targets in each step, thus generating an adaptive patrol path within the block.

[0045] It is important to note that, to avoid singular values ​​in the path selection probability due to the straight-line distance being close to zero when the robot is already very close to the device (i.e., the straight-line distance is less than the preset lower distance limit), the straight-line distance is assigned the preset lower distance limit. Specifically, if the number of candidate devices in the target block is equal to 1, the preset lower distance limit is assigned to 0.1 meters; if the number of candidate devices in the target block is greater than 1, the preset lower distance limit is set to one-tenth of the average distance between any two candidate devices in the target block.

[0046] Preferably, in some possible implementations of the embodiments of the present invention, the preset exit conditions include: the proportion of candidate devices in the target block is less than a preset lower limit for inspection, and the risk pheromones of all candidate devices are less than a preset lower limit for pheromones. It should be noted that the preset exit conditions indicate that most devices in the block have been inspected and the potential failure risk of uninspected devices is extremely low. This can ensure that high-risk devices are not missed, while avoiding excessive time spent by the robot on low-risk devices, thereby releasing resources in a timely manner and moving to the next target block.

[0047] In this embodiment of the invention, the preset lower limit for inspection is set as the empirical value of the proportion of uncovered equipment allowed in the historical routine inspection tasks of the substation, preferably set to 0.1; the preset lower limit for pheromones is determined before the start of each full-station inspection cycle by statistically analyzing the median of risk pheromones of equipment with normal inspection results in the previous complete inspection cycle.

[0048] Preferably, in some possible implementations of the embodiments of the present invention, the method for generating an inspection path and performing detection tasks on the inspection items along the inspection path includes: pre-constructing an inspection item association graph for different device types, including different inspection item nodes connected by directed edges, with the directed edges configured with initial path pheromones; using the source point of the inspection item association graph of the device type to which the current target device belongs as the initial current inspection node, and using the source point as the starting point of the target path; using the inspection item nodes pointed to by the directed edges starting from the current inspection node as candidate nodes of the current inspection node; multi-source inspection data including environmental data; determining the environmental state sensitivity based on the environmental data of the current target device's location at the current time; and determining the environmental state sensitivity based on the path pheromones of the directed edges pointing to each candidate node of the current inspection node and the current inspection... The system calculates the node jump weight of a candidate node based on its number of inspections and environmental sensitivity, and normalizes it to obtain the node jump probability. Based on the node jump weight, an inspection path is generated in the inspection item association graph corresponding to the current target device. Inspection tasks are then executed sequentially for each inspection item node along the inspection path. For each inspection item node on the inspection path, if the inspection result is abnormal, the number of times the inspection item node has not been selected into the inspection path since the last abnormality detection is obtained. The path pheromone of the directed edge pointing to the inspection item node is adjusted using this number, resulting in the adjusted path pheromone, thus updating the inspection item association graph corresponding to the current target device. The updated inspection item association graph is then used as the inspection item association graph for the current target device in the next inspection.

[0049] It should be noted that multiple inspection items of the equipment under inspection are obtained as inspection item nodes. These items include the body oil temperature, bushing oil level, cooler operating status, and on-load tap changer operation count for transformer-type equipment. Based on the progressive relationship of anomaly investigation among the inspection items, directed edges are established between related inspection item nodes. These directed edges point from the preceding inspection item node to the following inspection item node, representing the triggering of investigation of the following inspection item node when the preceding inspection item is abnormal. For example, when the detection result of the body oil temperature inspection item is determined to be abnormal, it triggers the tracing and investigation of the cooler operating status and oil chromatography online monitoring data. An initial path pheromone is configured for each directed edge. The directed acyclic graph formed by the inspection item nodes and directed edges serves as the inspection item association graph. Specifically, the initial path pheromone of all directed edges in the inspection item association graph is set to 1.

[0050] Multi-source inspection data also includes ambient temperature, ambient humidity, rainfall, water depth, and specific gas concentrations such as sulfur hexafluoride (SF6) gas leakage concentration. Each type of inspection item is pre-configured with its associated multi-source inspection data and its sensitivity threshold range; for example, the mechanism box condensation inspection item is associated with ambient humidity, with a sensitivity threshold range set to relative humidity greater than 80%; the SF6 circuit breaker liquefaction inspection item is associated with ambient temperature, with a sensitivity threshold range set to temperature below -2 degrees Celsius; the terminal box moisture inspection item is associated with rainfall and water depth, with sensitivity threshold ranges set to daily rainfall greater than 2 mm and water depth greater than 5 cm, respectively. Either trigger activates the environmental sensitivity of that inspection item. For each type of environmental data, if the environmental data does not fall within the preset sensitivity threshold range for that inspection item, it is determined that the current environment does not pose a sensitive impact on that inspection item, and the environmental state sensitivity is assigned a value of 1. If the environmental data falls within the sensitivity threshold range for that inspection item, it is determined that the current environmental conditions trigger a sensitivity for that inspection item, and the environmental state sensitivity is assigned an enhancement value greater than 1. Preferably, the enhancement value is set according to the sensitivity level, for example, a value of 2 for mild sensitivity, a value of 3 for moderate sensitivity, and a value of 5 for severe sensitivity. When multiple multi-source inspection data trigger sensitive conditions simultaneously, the environmental state sensitivity takes the largest enhancement value among them, achieving adaptive matching between the inspection checklist and the on-site environment.

[0051] To avoid a rigid pattern where high-value inspection items are repeatedly checked while less frequently checked items are ignored, the node jump weight in the inspection item association graph is determined by considering the following three factors: First, the accumulated path pheromone, representing the frequency of historical inspections where defects are discovered by delving along directed edges. A higher path pheromone value indicates a higher confidence level that a candidate node is a high-value discovery link, and the robot should be guided to prioritize reproducing that path. Therefore, path pheromone is positively correlated with node jump weight. Second, environmental state sensitivity, representing the degree to which the environmental conditions at the current target equipment location stimulate the probability of defect occurrence. A higher environmental state value indicates a more favorable environment for the manifestation of the corresponding fault mode, and the inspection should be prioritized. Therefore, environmental state sensitivity is positively correlated with node jump weight. Third, the number of times the same inspection item has been checked is used to suppress repeated checks of the same inspection item in a single inspection. A higher number of times indicates that the candidate node has been visited multiple times in this inspection, and continued repeated checks will crowd out the opportunity to investigate other insufficiently explored inspection items. Therefore, the number of times the same inspection item has been checked is negatively correlated with node jump weight. In this embodiment of the invention, the node jump weight of the candidate node is calculated as the product of the path pheromone of the directed edges pointing from the current inspection node to its candidate nodes, the environmental state sensitivity, and the reciprocal of the sum of the number of times the candidate node has been inspected in this inspection and a constant 1. The constant 1 is used to prevent the denominator from being meaningless if the candidate node has not yet been inspected in this inspection. The node jump probability of each candidate node is obtained by dividing the node jump weight of the current inspection node by the sum of the node jump weights of all candidate nodes, thus normalizing the node jump weights.

[0052] In this embodiment of the invention, the method for obtaining the inspection path includes: selecting candidate nodes from the candidate nodes of the current inspection node whose node jump probability is greater than or equal to a preset jump lower limit and recording them as valid jump nodes; if the current inspection node has a valid jump node, then using a roulette wheel algorithm based on the node jump probability, selecting an inspection item node from the valid jump nodes and adding it to the target path for updating, and using the selected inspection item node as the new current inspection node; if the current inspection node does not have a candidate node or a valid jump node, then backtracking to the previous inspection item node in the target path and using that inspection item node as the new current inspection node; until backtracking to the source point and the source point does not have a candidate node or a valid jump node, stopping the updating of the target path, and using the updated target path as the inspection path.

[0053] It should be noted that, to streamline the checklist and avoid wasting time on low-value checks, candidate nodes with a jump probability greater than or equal to a preset lower jump limit are designated as valid jump nodes. The jump probability represents the necessity of continuing the investigation along the current check item; the higher the value, the higher the potential failure risk and investigation value of the corresponding check item. While favoring high-risk check items, the randomness of exploration is preserved. If a valid jump node exists for the current check node, a node is selected based on the jump probability using a roulette wheel algorithm and added to the target path, becoming the new current check node for further investigation. To ensure that other valid branches can be traversed after the current investigation branch ends, if no candidate node or valid jump node exists for the current check node, the process backtracks to the previous check item node in the target path and uses it as the new current check node. The current check node that caused this backtracking is marked as an invalid node and removed from the candidate node set of the new current check node to prevent repeated entry into this invalid branch in subsequent roulette wheel selections. The above-mentioned in-depth and backtracking operations are continuously iterated until the backtracking reaches the source point of the association graph and the source point no longer has a valid jump node, at which point the update stops and the target path at this time is taken as the final inspection path of the device.

[0054] In this embodiment of the invention, the method for obtaining the preset jump lower limit is as follows: for each type of device, extract the historical inspection records of the most recent 90 days from the robot backend server, and filter out all complete records that contain the inspection item association graph search process; for each graph search, record the node jump probability value of each candidate node; and use the median of the jump probability values ​​of all candidate nodes in all records as the preset jump lower limit of this type of device in the current inspection cycle.

[0055] In this embodiment of the invention, if an anomaly is detected in a certain inspection item node during the current inspection, it means that the corresponding troubleshooting path has real defect discovery value. The more times an inspection item node has not been selected into an inspection path since the last anomaly was detected, the more it indicates that the high-value troubleshooting path has been neglected for a long time, and its path pheromone has significantly decayed due to natural evaporation or competitive inhibition, making it difficult to regain a sufficient probability of being selected in subsequent searches. Therefore, the sum of the constant 1 and the number of times the inspection item node has not been selected into an inspection path since the last anomaly was detected is calculated as an enhancement factor, and the product of this sum and the path pheromone of the directed edge pointing to the inspection item node is used as the adjusted path pheromone of the directed edge. This updates the path pheromone, allowing the long-neglected troubleshooting path to regain the opportunity to be selected in the next inspection's graph search, thereby effectively preventing the inspection list from permanently omitting certain defect patterns due to path dependence and ensuring the long-term adaptability of the graph. Specifically, the attenuation of path pheromones through natural evaporation or competitive suppression can be achieved as follows: after each inspection of the current target device, or at the end of each full-station inspection cycle, the path pheromones of all directed edges in the inspection item association graph are updated according to the formula P=P×(1-ρ), where P represents the path pheromone and ρ represents the attenuation coefficient between 0 and 1, the specific value of which is determined by the attenuation law of path pheromones; in this embodiment, 0.2 is used as an example. It should be noted that, as another implementation method, the above-mentioned path pheromone enhancement update process due to path pheromone attenuation may not be required.

[0056] Example 2: A mobile robot includes a memory and a processor, wherein the memory stores instructions; the processor invokes the instructions in the memory to cause the mobile robot to perform the steps of the substation autonomous inspection method based on multi-source information fusion provided in the above embodiments.

[0057] Example 3: This invention proposes a substation autonomous inspection system based on multi-source information fusion. Please refer to [link / reference]. Figure 3 The diagram illustrates a system architecture of a substation autonomous inspection system based on multi-source information fusion, according to an embodiment of the present invention. The system includes: The data acquisition module is used to acquire multi-source inspection data of each piece of equipment to be inspected in the substation and the real-time position of each mobile robot in real time, and to divide the inspection area of ​​the substation into different inspection zones. The inspection block generation module is used to update the risk pheromones of each device to be inspected based on multi-source inspection data; obtain the inspection demand degree of each inspection zone based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval; and merge the inspection zones into different inspection blocks. The target block selection module is used to select the target block of the mobile robot based on the inspection demand of the inspection area in each inspection block and the distance between the real-time position of the mobile robot and the inspection block, and control the mobile robot to move to the target block. The target device selection module is used to select the current target device of the mobile robot and proceed to inspect it based on the risk pheromone and spatial location characteristics of the uninspected devices in the target block; after completing the inspection of the current target device, the mobile robot selects a new current target device, and the inspection of the target block ends when the remaining uninspected devices in the target block meet the preset exit conditions, and a new target block is selected for the mobile robot. The inspection process for the current target equipment includes: generating an inspection path based on the historical inspection association features and environmental status features of the current target equipment, and performing detection tasks on the inspection items along the inspection path to complete the inspection of the current target equipment.

[0058] It should be noted that the devices provided in the above embodiments are only illustrative examples of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the multi-source information fusion substation autonomous inspection system and the multi-source information fusion substation autonomous inspection method embodiment provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

Claims

1. A substation autonomous inspection method based on multi-source information fusion, characterized in that, The method includes: The system acquires multi-source inspection data of each piece of equipment to be inspected in the substation in real time, as well as the real-time location of each mobile robot, and divides the inspection area of ​​the substation into different inspection zones. Based on multi-source inspection data, update the risk pheromones of each device to be inspected; based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval, obtain the inspection demand of each inspection zone, and merge the inspection zones into different inspection blocks. Based on the inspection demand of each inspection zone and the distance between the mobile robot's real-time position and the inspection zone, the target zone of the mobile robot is selected, and the mobile robot is controlled to move to the target zone. Based on the risk pheromones and spatial location characteristics of the uninspected devices in the target block, the mobile robot selects the current target device and proceeds to inspect it; after completing the inspection of the current target device, the mobile robot selects a new current target device, and the inspection of the target block ends when the remaining uninspected devices in the target block meet the preset exit conditions, and a new target block is selected for the mobile robot. The inspection process for the current target equipment includes: generating an inspection path based on the historical inspection association features and environmental status features of the current target equipment, and performing detection tasks on the inspection items along the inspection path to complete the inspection of the current target equipment.

2. The substation autonomous inspection method based on multi-source information fusion according to claim 1, characterized in that, The step of updating the risk pheromones of each device under inspection based on multi-source inspection data includes: Based on the life cycle stage of each device to be inspected, an initial risk pheromone is set for the corresponding device; The multi-source inspection data includes: different types of abnormal events of the equipment under inspection; For each device under inspection, when the multi-source inspection data of the device under inspection at the current moment contains an abnormal event, the time interval between the abnormal event of the device under inspection and the most recent abnormal event of the same type is obtained and recorded as the abnormal reproduction time; the abnormal reproduction time is negatively correlated and normalized, and the sum of the processing result and the constant 1 is used as the pheromone adjustment coefficient. When the multi-source inspection data of the device under inspection does not contain any abnormal events at the current moment and the device under inspection has completed the inspection, the number of inspections in which no abnormalities were found between the current moment and the most recent abnormal event is obtained, and the number of inspections is negatively correlated and normalized to obtain the pheromone adjustment coefficient. The risk pheromone of the device under inspection is multiplied by the pheromone adjustment coefficient to update the risk pheromone and obtain the updated risk pheromone of the device under inspection.

3. The substation autonomous inspection method based on multi-source information fusion according to claim 1, characterized in that, The step of obtaining the inspection demand level of each inspection zone and merging the inspection zones into different inspection blocks includes: For each inspection zone, the maximum value of the risk pheromone of all devices to be inspected within the inspection zone is extracted; the multi-source inspection data includes the historical inspection completion time of each device to be inspected; the time difference between the current time and the most recent inspection completion time of the inspection zone is taken as the historical inspection time interval; the product of the historical inspection time interval and the maximum value is calculated as the inspection demand degree of the inspection zone. The demand level of each inspection zone within the inspection area is determined based on the demand for inspection. Region growth is performed by taking any inspection zone within the inspection area as a growth point. Inspection zones with the same demand level as the growth point within the preset neighborhood of the growth point are taken as new growth points for region growth. Region growth stops when there are no inspection zones with the same demand level as the growth point within the preset neighborhood of all new growth points, resulting in different inspection blocks.

4. The substation autonomous inspection method based on multi-source information fusion according to claim 1, characterized in that, The selection of the target area for the mobile robot includes: The demand level includes a high demand level; the percentage of inspection zones at the high demand level in each inspection block is counted; the average value of the demand level of all inspection zones in each inspection block is calculated, and the sum of the percentage of the number and a constant 1 is multiplied by the average value to obtain the block urgency of the corresponding inspection block. Obtain the distance from the real-time position of the mobile robot to the geometric center of each inspection block, and use the ratio of the block urgency to the distance as the block selection weight for the corresponding inspection block; Based on the block selection weights of all inspection blocks, the block selection weights of each inspection block are normalized to obtain the block selection probability; based on the block selection probability, the roulette wheel algorithm is used to select the target block of the mobile robot from all inspection blocks.

5. The substation autonomous inspection method based on multi-source information fusion according to claim 1, characterized in that, The selection of the current target device for the mobile robot includes: Record the uninspected devices within the target block of the mobile robot as candidate devices at the current moment; obtain the shortest distance from each candidate device within the target block to the boundary of the target block, and the straight-line distance from the real-time position of the mobile robot to each candidate device at the current moment. Based on the risk pheromone of each candidate device, the shortest distance, and the straight-line distance, the device candidate weight of the corresponding candidate device is obtained; the risk pheromone and the shortest distance are both positively correlated with the device candidate weight, and the straight-line distance is negatively correlated with the device candidate weight; Based on the device candidate weights of all candidate devices within the target block, the device candidate weights of each candidate device are normalized to obtain the device selection probability; based on the device selection probability, the roulette wheel algorithm is used to select the current target device from all candidate devices.

6. The substation autonomous inspection method based on multi-source information fusion according to claim 1, characterized in that, The process of generating an inspection path and performing detection tasks on the inspection items along the inspection path includes: A pre-constructed association graph of inspection items for different device types is constructed, including different inspection item nodes connected by directed edges, wherein the directed edges are configured with initial path pheromones; The source node of the inspection item association graph of the current target device type is used as the initial current inspection node, and the source node is used as the starting point of the target path; the inspection item node pointed to by the directed edge starting from the current inspection node is used as the candidate node of the current inspection node. Multi-source inspection data includes environmental data; environmental state sensitivity is determined based on the environmental data at the current time of the current target device's location; the node jump weight of the candidate node is obtained and normalized according to the path pheromone of the directed edges pointing to each candidate node of the current inspection node, the number of times the candidate node has been inspected in this inspection, and the environmental state sensitivity, so as to obtain the node jump probability. Based on the node jump weight, an inspection path is generated in the inspection item association graph corresponding to the current target device; The inspection task is executed sequentially for each inspection item node along the inspection path. For each inspection item node on the inspection path, if the inspection result of the inspection item node is abnormal, the number of times the inspection item node has not been selected into the inspection path since the last abnormality was detected is obtained. The path pheromone of the directed edge pointing to the inspection item node is adjusted using the number of times, and the adjusted path pheromone of the directed edge is obtained to update the inspection item association graph corresponding to the current target device. The updated inspection item association graph is then used as the inspection item association graph of the current target device in the next inspection.

7. The substation autonomous inspection method based on multi-source information fusion according to claim 6, characterized in that, The generation of the inspection path in the inspection item association graph corresponding to the current target device based on node jump weight includes: Select candidate nodes from the current check node whose node jump probability is greater than or equal to the preset jump lower limit and record them as valid jump nodes; If there is a valid jump node at the current inspection node, a check item node is selected from the valid jump nodes based on the node jump probability and added to the target path for updating, and the selected check item node is used as the new current inspection node. If there are no candidate nodes or valid jump nodes at the current inspection node, then backtrack to the previous inspection item node in the target path and use that inspection item node as the new current inspection node. The update of the target path stops when the path is traced back to the source point and there are no candidate nodes or valid jump nodes at the source point. The updated target path is then used as the inspection path.

8. The substation autonomous inspection method based on multi-source information fusion according to claim 5, characterized in that, The preset exit conditions include: the proportion of the number of candidate devices in the target block is less than the preset lower limit for inspection, and the risk pheromones of all candidate devices are less than the preset lower limit for pheromones.

9. A mobile robot, characterized in that, The mobile robot includes a memory and a processor, wherein the memory stores instructions; the processor invokes the instructions in the memory to cause the mobile robot to perform the steps of the substation autonomous inspection method based on multi-source information fusion as described in any one of claims 1 to 8.

10. A substation autonomous inspection system based on multi-source information fusion, characterized in that: The system includes: The data acquisition module is used to acquire multi-source inspection data of each piece of equipment to be inspected in the substation and the real-time position of each mobile robot in real time, and to divide the inspection area of ​​the substation into different inspection zones. The inspection block generation module is used to update the risk pheromones of each device to be inspected based on multi-source inspection data; obtain the inspection demand degree of each inspection zone based on the risk pheromones of the devices to be inspected in each inspection zone and the historical inspection time interval; and merge the inspection zones into different inspection blocks. The target block selection module is used to select the target block of the mobile robot based on the inspection demand of the inspection area in each inspection block and the distance between the real-time position of the mobile robot and the inspection block, and control the mobile robot to move to the target block. The target device selection module is used to select the current target device of the mobile robot and proceed to inspect it based on the risk pheromone and spatial location characteristics of the uninspected devices in the target block; after completing the inspection of the current target device, the mobile robot selects a new current target device, and the inspection of the target block ends when the remaining uninspected devices in the target block meet the preset exit conditions, and a new target block is selected for the mobile robot. The inspection process for the current target equipment includes: generating an inspection path based on the historical inspection association features and environmental status features of the current target equipment, and performing detection tasks on the inspection items along the inspection path to complete the inspection of the current target equipment.

Images