A task allocation method and system for a maintenance oil depot inspection robot based on an internet of things

Abstract

The application belongs to the technical field of robot control and automation task management, and relates to a task allocation method and system for a maintenance oil depot inspection robot based on the Internet of Things, which comprises the following steps: constructing a dynamic risk situation semantic model based on an oil depot topology and real-time environment parameters; identifying a composite abnormal event through logical reasoning by fusing multi-modal data; calculating a task instant risk utility score in combination with the model; generating an elastic resource vector by collecting a robot operating state and communication characteristics; solving an optimization function to generate a deployment instruction stream by taking the score as a weight and the vector as a constraint; and driving a job according to the instruction stream and performing self-adaptive correction based on feedback. The application solves the problem that the existing task allocation mode of an inspection robot has deficiencies in fusing multi-dimensional dynamic risk situations and robot elastic resource constraints, resulting in a lack of adaptability and reliability of task deployment of the system in a complex environment.

Description

Technical Field

[0001] This invention belongs to the technical field of robot control and automated task management, and relates to a task allocation method and system for an oil depot inspection robot based on the Internet of Things. Background Technology

[0002] In high-risk industrial environments such as engine room and oil depots, routine equipment inspections are a fundamental aspect of ensuring safe production. These environments are typically characterized by complex spatial structures and dynamically changing distributions of potential hazards. To address these challenges, automated inspection robots have been introduced to replace or assist manual inspections, aiming to reduce personnel safety risks and increase the coverage and frequency of inspections.

[0003] Currently, the most commonly used solutions in the industry mainly include robot inspection systems based on fixed path planning. In this type of solution, the inspection robot typically performs tasks according to a pre-set schedule and route, and its behavioral logic is relatively fixed. Some systems integrate single-point sensor alarm functions, which may trigger the robot to investigate when a fixed sensor detects an anomaly. These systems usually decouple the robot's task scheduling from the overall environmental situational awareness network of the oil depot, and the robot, as an independent execution unit, has relatively limited decision-making basis.

[0004] However, the aforementioned traditional methods have certain limitations when dealing with complex and dynamic scenarios. First, inspection strategies based on fixed schedules and paths lack adaptability to real-time risk dynamics. For example, when the risk level of a certain area increases due to minor leaks or specific wind directions, the fixed inspection plan may not be able to adjust resources in time, resulting in insufficient attention to potentially high-risk areas. Second, existing task triggering mechanisms typically rely on threshold alarms from single sensors. Their ability to identify complex fault precursors caused by multiple concurrent weak abnormal signals is limited, potentially leading to delayed responses. Furthermore, the task allocation process often fails to fully consider the robot's own dynamic resource status, such as remaining battery power, sensor health, and real-time communication quality at its location. This can lead to unreasonable task allocation, posing a risk of task failure or failure to transmit critical data. Summary of the Invention

[0005] In a first aspect, the present invention provides a task allocation method for an oil depot inspection robot based on the Internet of Things, comprising the following steps: S1. Based on the topological information and real-time environmental parameters of the target oil depot, construct a dynamic risk situation semantic model; S2. Integrates multimodal heterogeneous data from inspection robots and fixed sensors to identify complex abnormal event entities through logical reasoning; S3. Combine the dynamic risk situation semantic model to calculate the instantaneous risk utility score of each task in the real-time task pool. S4. Collect the operating status and current location communication characteristics of the inspection robot to generate a robot elastic resource vector; S5. Using the immediate risk utility score as the weight and the robot's elastic resource vector as the constraint, solve the optimization function to generate the task allocation instruction flow; S6. Drive the robot to operate according to the task allocation instruction flow, and adaptively correct the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

[0006] A further aspect of the present invention, step S1, includes the following steps: Obtain a 3D topological map of the target oil depot, define the explosion-proof level of each functional area and the physical spatial coordinates of the equipment through attribute annotation, and form a static semantic basis; Real-time monitoring data from fixed IoT sensors is retrieved to obtain real-time environmental parameters, including ambient temperature and humidity, wind direction, and concentration of characteristic gases. By integrating real-time environmental parameters into a static semantic foundation, and updating the physical facility status and mapping dynamic security boundaries, a dynamic risk situation semantic model is generated.

[0007] A further aspect of the present invention, step S2, includes the following steps: Perform spatiotemporal correlation analysis on multimodal heterogeneous data to extract sensor data clusters that are within a preset time window and whose spatial distance is less than a preset distance threshold; By using a chain-based reasoning model combined with a pre-set fault feature library, the causal relationship between different sensor physical indicators in the data cluster is determined. Based on the judgment results, the root cause failure mode is identified, and the scope of risk impact and urgency level rating are determined by combining the dynamic risk situation semantic model, generating composite abnormal event entities.

[0008] A further aspect of the present invention, step S3, includes the following steps: A risk propagation mathematical model is constructed, and the edge weights of the model are dynamically adjusted based on the wind flow direction in the dynamic risk situation semantic model to calculate the risk spread weight of each task. Extract the signal strength index of the region corresponding to each task, and convert the signal strength index into the probability of communication feasibility through a mapping function; The risk contagion weight is weighted by the communication feasibility probability to obtain the immediate risk utility score for each task.

[0009] A further aspect of the present invention, step S4, includes the following steps: The system collects real-time data on the robot's remaining battery percentage, instantaneous power consumption rate, drive motor load status, and the online health status of onboard sensors to quantify hardware performance capabilities. Based on the current coordinates of the inspection robot, the dynamic risk situation semantic model is queried, and the available bandwidth of the current environment is predicted by combining historical network throughput data. The hardware execution capability indicators and the predicted available bandwidth are structured and encapsulated to generate a robot elastic resource vector.

[0010] A further aspect of the present invention, step S5, includes the following steps: An optimization function is constructed with the goal of increasing the total weight score of tasks completed per unit time, and the robot's elastic resource vector is transformed into dynamic boundary constraints of power, hardware health and communication capabilities. By employing a multi-objective heuristic algorithm, an execution scheme is optimized that includes the sequence of movement path points, sensor activation periods, and data sampling density, while satisfying the access rules for the explosion-proof zones of the oil depot. The execution plan is encoded as a task dispatch instruction flow that controls the physical behavior of the robot.

[0011] A further aspect of the present invention is to monitor the real-time communication quality during the robot's execution process. When the communication quality is lower than a preset bandwidth threshold, the data transmission granularity is adjusted and fed back to step S3 to correct the communication feasibility probability of subsequent tasks. Identify the updated equipment status and environmental parameters during the inspection task execution process, and feed them back to step S1 to correct the attribute status and safety boundary distribution in the dynamic risk situation semantic model.

[0012] A further aspect of the present invention includes the following steps for adjusting the data transmission granularity: Reduce the data stream frame rate and image resolution of the vision camera, and activate the local storage backup strategy to cache non-urgent data to the robot's built-in storage medium, and upload it in batches after the communication quality is restored.

[0013] A further aspect of the present invention includes the following steps for determining causal relationships: The data clusters are input into a Bayesian network model based on a directed acyclic graph, and the posterior probability of each potential fault event is calculated. When the posterior probability exceeds a preset decision threshold, it is determined that there is a causal relationship between the sensor physical indicators.

[0014] Secondly, the present invention provides a task allocation system for an oil depot inspection robot based on the Internet of Things, comprising the following modules: The risk situation semantic modeling module constructs a dynamic risk situation semantic model based on the topological information and real-time environmental parameters of the target oil depot. The composite anomaly diagnosis module is used to integrate multimodal heterogeneous data from inspection robots and fixed sensors, and identify composite anomaly entities through logical reasoning. The task risk utility calculation module is used to calculate the real-time risk utility score of each task in the real-time task pool by combining the dynamic risk situation semantic model. The robot resource vectorization module is used to collect the operating status and current location communication characteristics of the inspection robot and generate the robot's elastic resource vector. The task allocation instruction generation module is used to solve an optimization function to generate a task allocation instruction flow by using the real-time risk utility score as the weight and the robot's elastic resource vector as the constraint. The closed-loop strategy adaptive module drives the robot to operate according to the task allocation instruction flow, and adaptively corrects the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

[0015] In summary, the present invention has the following beneficial technical effects: 1. By constructing a dynamic risk situation semantic model that integrates 3D topological information and real-time environmental parameters, this invention can transform static geographic space into a digital scene containing risk dimensions. Based on this model, the instantaneous risk utility score of each task is calculated, changing the traditional scheduling mode that relies on fixed plans. This allows inspection resources to be dynamically allocated according to changes in the actual risk level, which helps to improve the monitoring coverage of high-risk areas.

[0016] 2. By introducing a robot elastic resource vector, factors such as the inspection robot's power status, hardware health, and real-time communication bandwidth at its location are quantified as dynamic constraints for task allocation. When generating execution plans, the optimization algorithm can pre-assess the robot's actual capability boundaries, thereby avoiding execution interruptions caused by insufficient power, hardware damage, or weak communication signals. This helps reduce the probability of task failure and ensures the continuity of inspection operations.

[0017] 3. This invention achieves adaptive correction of model parameters by real-time monitoring of communication quality and environmental parameter changes during task execution and feeding the measured data back to the risk model and task allocation module. This closed-loop structure supports the system in continuously optimizing subsequent tasks based on actual on-site conditions, effectively addressing dynamic changes in the oil depot's internal environment and enhancing the robustness and adaptability of the entire task allocation system in complex industrial scenarios. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0019] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0020] Figure 2 Structural schematic diagrams of embodiments of this application are disclosed. Detailed Implementation

[0021] The following is in conjunction with the appendix Figure 1 - Figure 2 A preferred description of the present invention is provided below.

[0022] See attached document Figure 1 This invention proposes a task allocation method for an IoT-based oil depot inspection robot, comprising the following steps: S1. Based on the topological information and real-time environmental parameters of the target oil depot, construct a dynamic risk situation semantic model; S2. Integrates multimodal heterogeneous data from inspection robots and fixed sensors to identify complex abnormal event entities through logical reasoning; S3. Combine the dynamic risk situation semantic model to calculate the instantaneous risk utility score of each task in the real-time task pool. S4. Collect the operating status and current location communication characteristics of the inspection robot to generate a robot elastic resource vector; S5. Using the immediate risk utility score as the weight and the robot's elastic resource vector as the constraint, solve the optimization function to generate the task allocation instruction flow; S6. Drive the robot to operate according to the task allocation instruction flow, and adaptively correct the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

[0023] In one embodiment of the present invention, step S1 includes the following steps: Obtain a 3D topological map of the target oil depot, define the explosion-proof level of each functional area and the physical spatial coordinates of the equipment through attribute annotation, and form a static semantic basis; Real-time monitoring data from fixed IoT sensors is retrieved to obtain real-time environmental parameters, including ambient temperature and humidity, wind direction, and concentration of characteristic gases. By integrating real-time environmental parameters into a static semantic foundation, and updating the physical facility status and mapping dynamic security boundaries, a dynamic risk situation semantic model is generated.

[0024] First, a comprehensive three-dimensional spatial data acquisition of the target oil depot area is conducted by deploying a mobile LiDAR scanning system and a high-resolution visual imaging system. This system acquires point cloud data and corresponding image sequences with preset linear and angular accuracies, such as a linear accuracy of 0.02 m and an angular accuracy of 0.05°. The point cloud data is processed in real-time using a Simultaneous Localization and Mapping (SLAM) algorithm to generate a high-precision three-dimensional topological map of the target oil depot. This high-precision three-dimensional topological map is a digital model of the oil depot with accurate geometric shape and spatial coordinates, generated using LiDAR and a high-resolution visual system, and generated through SLAM algorithms and 3D reconstruction technology. Its linear accuracy is typically better than 0.05 m.

[0025] The high-precision 3D topological map is texture-mapped and gridded reconstructed based on a geographic coordinate system and stored as a renderable grid model. Subsequently, using the equipment list and area division information in the oil depot management system, interactive attribute annotation is performed on the high-precision 3D topological map using Geographic Information System (GIS) software. Specifically, for each functional area within the oil depot, such as tank areas, loading and unloading areas, and pipeline corridors, a corresponding explosion-proof level is assigned, for example, Zone 0, Zone 1, and Zone 2, and this explosion-proof level is embedded as an area attribute into the map data structure. It should be understood that the explosion-proof level is a regional explosion hazard level classified according to industrial explosion hazard environment standards. For example, Zone 0 refers to an area where an explosive gas atmosphere is continuously present or exists for a long time; Zone 1 refers to an area where an explosive gas atmosphere may occur during normal operation; and Zone 2 refers to an area where an explosive gas atmosphere is unlikely to occur during normal operation.

[0026] Simultaneously, for various equipment within the oil depot, such as storage tanks, valve assemblies, and electrical equipment, their precise physical spatial coordinates in the 3D topology map are associated and stored by reading their unique asset IDs, for example, XYZ coordinates. Physical spatial coordinates represent the precise location information of equipment or areas within the oil depot in a specific coordinate system, such as WGS84 or a local Cartesian coordinate system, with an accuracy better than 0.1m. Furthermore, structured data such as equipment type, material, design pressure, design temperature, and last maintenance date also need to be associated and stored. The high-precision 3D topology map, containing explosion-proof ratings, equipment physical spatial coordinates, and other structured attribute information, together constitute the static semantic foundation. In other words, the static semantic foundation is the high-precision 3D topology map and its associated, non-rapidly changing set of structured information. The static semantic foundation is stored in a graph database, such as Neo4j, where nodes represent areas or equipment entities, edges represent spatial adjacency or functional relationships, and attributes store detailed information such as explosion-proof ratings and coordinates.

[0027] The system retrieves real-time monitoring data from fixed IoT sensors deployed within the oil depot area via a wireless communication network, acquiring real-time environmental parameters including ambient temperature and humidity, wind direction, and characteristic gas concentration. In this embodiment, the real-time environmental parameters are timestamped environmental element monitoring data streams from the fixed IoT sensors, specifically including ambient temperature and humidity, wind direction, and characteristic gas concentration. Ambient temperature and humidity refer to the real-time atmospheric temperature and relative humidity measured by the sensors; wind direction refers to the real-time wind speed and direction acquired by the sensors; and characteristic gas concentration refers to the volume percentage or ppm concentration of a specific flammable, explosive, or toxic gas detected by the sensors in the environment.

[0028] The fixed IoT sensors include a temperature sensor, a humidity sensor, an ultrasonic anemometer and wind direction sensor, and a multi-channel gas detector. In this embodiment, the temperature sensor has a range of -40℃ to 80℃ and an accuracy of ±0.5℃; the humidity sensor has a range of 0%RH to 100%RH and an accuracy of ±3%RH; the ultrasonic anemometer and wind direction sensor has a wind speed range of 0 m / s to 60 m / s and an accuracy of ±0.5 m / s; the wind direction accuracy is ±5°; and the multi-channel gas detector can detect characteristic gases such as methane, hydrogen sulfide, and carbon monoxide, with a range of 0 ppm to 1000 ppm and an accuracy of ±5 ppm. Sensor data is collected via an edge gateway using the MQTT protocol. The sensor ID, timestamp, and measurement values ​​are uploaded to the central IoT platform in JSON format every second. It should be noted that the data acquisition time interval... The time interval is typically set from 5 to 60 seconds, depending on the sensor type and the importance of the data. The central IoT platform writes the data to a high-speed time-series database, such as InfluxDB. The real-time environmental parameter data stream includes at least... Each data frame contains one or more sensor IDs, the corresponding geographic location, a timestamp, and the specific measurement value.

[0029] Real-time environmental parameters are integrated into a static semantic foundation to generate a dynamic risk situation semantic model. The specific integration process is as follows: First, a data fusion service, for example, using the Apache Flink stream processing engine, reads real-time environmental parameter data streams from a time-series database and continuously retrieves static semantic foundation data from a graph database. For each real-time environmental parameter data point, its corresponding physical space coordinates and nearby area and equipment information are queried from the static semantic foundation based on its sensor ID. Then, the physical facility status is updated according to preset rules. The physical facility status is an attribute in the dynamic risk situation semantic model that describes the operational health status of various equipment within the oil depot, such as normal, potential leak, overheat warning, etc. The specific judgment logic is as follows: For example, if a gas detector detects a methane concentration near a storage tank... continuous The second highest value was above the set threshold. If this happens, the status attribute of the storage tank entity in the graph database will be updated from normal to potential leak. In this embodiment, the methane concentration threshold of the gas detector... Typical value set at 50 ppm; number of consecutive detections Set to 3 times to avoid false alarms caused by occasional instantaneous fluctuations in readings.

[0030] Simultaneously, based on wind direction information and considering the tank's geometry and gas diffusion characteristics, a Gaussian plume model is executed. For example, simplified models such as AERMOD or CALPUFF are used to calculate potential gas diffusion areas and their risk concentration contour lines. This model considers factors such as wind speed, wind direction, ambient temperature, and atmospheric stability. The gas diffusion area is represented as a dynamic spatial polygon on a three-dimensional topological map. The dynamic risk situation is assessed according to the following rule: if the concentration of any characteristic gas within the area exceeds... or equipment temperature With ambient temperature The difference Exceed If so, the risk level of the area is increased. Specifically, the difference between the equipment temperature and the ambient temperature... The temperature is set at 15°C, based on the normal operating thermal characteristics of typical equipment in an oil depot. Exceeding this value indicates a possible localized overheating anomaly. The risk gas concentration threshold is set at 10% LEL, or 0.5% of the methane volume concentration. This value is determined according to industrial safety regulations as the low-limit explosion hazard concentration requiring countermeasures. Dynamic risk status refers to the risk level or risk indicator calculated over time for a specific area or equipment within the oil depot based on real-time environmental parameters and preset rules, such as low risk, medium risk, and high risk.

[0031] Finally, based on the gas diffusion region generated by the Gaussian plume model and combined with the explosion-proof level defined in the static semantic foundation, the dynamic safety boundary distribution is dynamically calculated. For example, if the predicted gas concentration at any point in the gas diffusion region reaches... The explosion-proof level of the area where that point is located will be temporarily upgraded to [level]. The area is defined, and its boundaries are updated in real time on a 3D topological map to form a dynamic safety boundary distribution. In this embodiment, the explosion risk level is increased. A value of 1 indicates that when the predicted gas concentration reaches a specific percentage of the lower explosive limit (LEL), the corresponding explosion-proof zone level will be raised by at least one level. Dynamic safety boundary distribution is the area boundary that is dynamically adjusted and displayed on a 3D topology map based on real-time environmental parameters and explosion-proof levels, used to indicate temporary restricted or warning areas.

[0032] All the integrated dynamic information, such as the updated physical facility status, dynamic risk level, and geometric description of the dynamic safety boundary, is written into the graph database in real time as new attributes and relationships, collectively forming a dynamic risk situation semantic model. This model is a real-time updated digital high-dimensional information model of the oil depot, integrating static semantic foundations, real-time environmental parameters, physical facility status, dynamic risk situation, and dynamic safety boundary distribution, used to support risk assessment and task allocation decisions; the dynamic risk situation semantic model is continuously updated every [period]. Iterates once per second.

[0033] For example, suppose the current target oil depot's tank area is designated TA01, and its explosion-proof level is marked as Zone 2 in the static semantic basis. Within this area, a methane gas detector with sensor ID GS001 and a wind speed and direction sensor with ID WS001 are deployed. First, a mobile lidar scanning system acquires centimeter-level precision 3D point cloud data of this area, and after SLAM processing, generates an entity containing TA01 and its internal tanks. Valve group A high-precision 3D topological map was created. Using GIS software, the outline of TA01 was accurately drawn on this map, and its explosion-proof level was marked as Zone 2, while also being associated with the storage tank. The physical spatial coordinates are (121.498, 31.229, 15.5) m. Subsequently, the IoT platform receives real-time methane concentration data from GS001 and real-time wind speed and direction data from WS001 every 10 seconds. For example, at t=0, GS001 reports a methane concentration of 10 ppm, WS001 reports a wind speed of 2.5 m / s, and a wind direction of 225° southwest. These data are considered the first batch of real-time environmental parameters. This methane concentration value is then compared with... A comparison revealed that its value was below the threshold; the storage tank... The physical facilities remained in normal condition. Then, at t=10s, GS001 reported that the methane concentration had increased to 55 ppm. Since 55 ppm is higher than... At 50 ppm, the data fusion service identified an abnormal methane concentration. (Storage tank) The physical facility status has been updated to indicate a potential leak.

[0034] Meanwhile, the service, based on the latest wind speed and direction reported by WS001, combined with the storage tank... Using the geographical location as a potential leak source, a Gaussian plume model was executed. The model calculated that, under the given wind direction and speed conditions, the potential methane diffusion area primarily covers the northwest direction of TA01, and extends to a distance from the storage tank. At approximately 20 m, the predicted methane concentration contour lines reach Based on this prediction, the service upgrades the dynamic risk situation of TA01 to medium risk - local gas diffusion. Simultaneously, since the predicted local methane concentration in the diffusion area has reached 10% LEL, the dynamic safety boundary distribution is dynamically calculated. The geometric boundary of the northwestern part of TA01, originally classified as Zone 2 explosion protection, is temporarily upgraded to Zone 1 on the corresponding 3D topological map and marked as a restricted access area. This is updated in real-time to the graph database as part of the dynamic risk situation semantic model.

[0035] In one embodiment of the present invention, step S2 includes the following steps: Perform spatiotemporal correlation analysis on multimodal heterogeneous data to extract sensor data clusters that are within a preset time window and whose spatial distance is less than a preset distance threshold; By using a chain-based reasoning model combined with a pre-set fault feature library, the causal relationship between different sensor physical indicators in the data cluster is determined. Based on the judgment results, the root cause failure mode is identified, and the scope of risk impact and urgency level rating are determined by combining the dynamic risk situation semantic model, generating composite abnormal event entities.

[0036] First, data from the vehicle-mounted sensors on the inspection robot and monitoring data from fixed IoT sensors are collected synchronously via a wireless communication link, forming a multimodal heterogeneous data stream. This multimodal heterogeneous data stream refers to a data set with diverse data formats and characteristics, aggregated through time synchronization from sensor data of different types (e.g., gas, temperature, vision, vibration) and different physical sources (e.g., fixed sensors and robot-mounted sensors). The vehicle-mounted sensors on the inspection robot include: a vehicle-mounted gas detector, an infrared thermal imager, an ultrasonic ranging sensor, and a vision camera. In this embodiment, the vehicle-mounted gas detector can be an electrochemical or catalytic combustion sensor with a detection accuracy of 1 ppm; the infrared thermal imager can be selected with a resolution of 640x480 pixels, a temperature measurement range of -20℃ to 550℃, and an accuracy of ±2℃; the ultrasonic ranging sensor can be selected with a range of 0.1 m to 5 m and an accuracy of ±0.01 m; and the vision camera can be selected with a resolution of 1920x1080 pixels and a frame rate of 25 fps.

[0037] The inspection robot encapsulates the collected data into timestamp-synchronized data packets via a standardized communication interface and transmits them to the edge computing node asynchronously via industrial-grade Wi-Fi or a 5G NR private network. Monitoring data from fixed IoT sensors, specifically the real-time environmental parameters mentioned in S1, including ambient temperature and humidity, wind direction, and characteristic gas concentrations, are also transmitted to the same edge computing node via a similar wireless communication link. Upon receiving data from both sensor types, the edge computing node first performs timestamp synchronization, aligning the measurement data from different sources to a unified time base. Then, it aggregates the data according to a preset data structure, such as a Protobuf-formatted serialized message, generating a multimodal heterogeneous data stream. In this embodiment, the average data transmission rate of the multimodal heterogeneous data stream is 5 Mbps to 20 Mbps, depending on the number of sensors and the sampling frequency.

[0038] Secondly, spatiotemporal correlation analysis is performed on the multimodal heterogeneous data stream, using a chain-inference model combined with a pre-defined fault feature library to determine the causal relationships between different sensor physical indicators. It should be understood that spatiotemporal correlation analysis is a data processing method used to identify heterogeneous sensor data points that are geographically and temporally close and statistically correlated. The spatiotemporal correlation analysis is performed as follows: For each data frame in the multimodal heterogeneous data stream, its precise geographic coordinates and timestamps are first extracted from the metadata associated with each sensor; for the vehicle-mounted sensors of the inspection robot, their geographic coordinates are provided by the robot's real-time RTK-GNSS positioning system, with an accuracy better than 0.05 m; for fixed sensors, their coordinates are preset fixed values. Then, based on the constructed dynamic risk situation semantic model, all data points within a specific time window are analyzed. And the spatial distance is less than a certain threshold The sensor data is clustered. In this embodiment, the time window... This is used to define the maximum value of the data time difference considered in spatiotemporal correlation analysis, preferably ranging from 5 to 20 seconds; spatial distance threshold. This is used to define the maximum value of the spatial distance of data considered in spatiotemporal correlation analysis, and its preferred range is 2 meters to 10 meters; for example, It can be set to 10 seconds. It can be set to 5m. During the clustering process, Euclidean distance is used as the spatial distance metric, and time difference is used as the temporal distance metric.

[0039] If the clustering results contain multiple heterogeneous sensor data points, these data are input into the chained inference model. The chained inference model is a method that uses a predefined causal relationship graph, such as a Bayesian network or rule-based reasoning system, to analyze observed anomalies to identify their potential root causes of faults. This model is constructed using a Bayesian network model based on a directed acyclic graph (DAG), where nodes represent different sensor physical indicators or potential fault events, and directed edges between nodes represent causal relationships. The conditional probability table of the Bayesian network is pre-trained by analyzing historical fault data and domain expert knowledge. The chained inference model then makes judgments based on a pre-defined fault feature library.

[0040] The fault feature library is a structured knowledge base containing common fault modes of various equipment within the oil depot, corresponding combinations of abnormal sensor features. For example, the fault mode of valve seal failure corresponds to feature combinations such as abnormal local temperature rise, increased methane concentration, and abnormal vibration, as well as the potential causal relationships between these feature combinations. When spatiotemporal correlation analysis identifies highly correlated heterogeneous data sets, the chain-inference model calculates the posterior probability of each potential fault event based on the causal rules defined in the fault feature library. For example, if infrared thermal imager data around a valve shows a continuous temperature increase, while an onboard gas detector detects an abnormal increase in methane concentration near the valve, and a fixed vibration sensor shows an abnormal vibration frequency, the chain-inference model will calculate the posterior probability of the valve seal failure fault mode using its constructed Bayesian network. If this posterior probability exceeds a preset decision threshold... If the model determines that there is a causal relationship between these physical indicators, it indicates the corresponding root cause failure mode. Decision threshold. It is the minimum posterior probability required for the chain reasoning model to determine the causal relationship. Its setting range is usually 0.7 to 0.9, and is adjusted according to the historical false alarm rate and false negative rate.

[0041] Finally, based on the judgment results, the root cause failure mode of the anomaly is identified, generating a composite anomaly event entity containing the anomaly event type, risk impact scope, and urgency rating. The composite anomaly event entity is a standardized data structure used to describe a diagnosed failure event caused by multiple interrelated anomalies, containing key information such as type, impact scope, and urgency. If the chain reasoning model determines that a causal relationship exists, and its posterior probability exceeds the decision threshold... Then, the root cause failure mode that best matches the judgment result is extracted from the failure feature library. The root cause failure mode will be directly used as the anomalous event type of the composite anomalous event entity. This attribute describes the nature of the event, such as valve seal failure or pipeline fatigue cracking. Further utilizing the constructed dynamic risk situation semantic model, combined with the physical location of the root cause failure mode, the risk propagation path and impact range in the current environment are analyzed to determine the risk impact range of the composite anomalous event entity. This range is the geographic spatial area that may lead to potential hazards, and its geometric description comes from the dynamic risk situation semantic model. For example, if the root cause failure mode is a storage tank... For small-scale valve leakage, the geometric boundary of the potential impact area of ​​gas diffusion is calculated based on real-time airflow direction and Gaussian plume model, which serves as the risk impact range.

[0042] The urgency rating is calculated by weighting the factors of the abnormal event type, the scope of its impact, and the risk level of the current local environment provided by the dynamic risk situation semantic model. The urgency rating classifies complex abnormal events into three levels: low, medium, and high, based on the event's potential harm, propagation speed, and impact on the safety of oil depot operations. The complex abnormal event entity is output as a structured JSON object.

[0043] For example, the detection of the storage tank is continued. A scenario involving abnormal methane concentrations in the area. At t=10s, the fixed IoT sensor GS001 reported a methane concentration of 55 ppm, which was flagged as a potential leak by S1. Subsequently, the storage tank... The dynamic risk situation in the area has been upgraded to medium risk – localized gas diffusion, and the explosion-proof level in a certain area northwest of TA01 has been temporarily upgraded to Zone 1. At this time, the inspection robot was dispatched to this medium-risk area for inspection. The inspection robot approached the storage tank. When a certain valve is inspected, its onboard sensors simultaneously collect and transmit data. Specifically, at t=20s, the onboard gas detector of the inspection robot is located at a distance of [distance missing] from the storage tank. valve A local methane concentration of 60 ppm was detected at the site.

[0044] Meanwhile, the infrared thermal imager scans the valve. The surface temperature was detected as 75°C, while the ambient temperature recorded in S1 for the same area was 25.3°C, resulting in a temperature difference of 49.7°C, significantly exceeding the set threshold of 15°C. Furthermore, the vibration sensor on the robot also detected the valve. High-frequency abnormal vibrations were observed at the location, with the root mean square value exceeding the normal baseline by three times. The geographic coordinates of the vehicle-mounted sensors were determined by the robot's RTK-GNSS system, and the distance from the abnormal point measured by GS001 was less than [missing information]. 5 m, and the timestamps are all Within 10 seconds, the edge computing node receives the data from the vehicle-mounted sensors and the 55 ppm methane concentration data from the GS001, then performs spatiotemporal correlation analysis. Because the multimodal data of localized high methane concentration, localized high temperature, and localized abnormal vibration all spatially point to the valve... Furthermore, these events occur synchronously in time, and the data is input into chain-based inference models, such as Bayesian networks.

[0045] The chain-reasoning model, combined with valve seal failure rules from a fault feature database, determines that the posterior probability of valve seal failure increases significantly when local methane concentration consistently exceeds 60 ppm, surface temperature is more than 40°C above ambient temperature, and abnormal vibration signals are present. Calculations show the model yields a posterior probability of 0.85 for valve seal failure, which is greater than the decision threshold. The value is 0.75. Based on this determination, the identified root cause failure mode is the valve. Seal failure. Therefore, a composite abnormal event entity is generated, with its abnormal event type set as: storage tank. valve Seal failure. The scope of the risk impact is determined based on the dynamic risk situation semantic model, combined with the valve... Based on the precise location and the current 240° wind direction of 2.8 m / s, it is estimated that the valve... Centered on the target area, a fan-shaped region with a radius of 30 m is temporarily upgraded to Zone 0 for explosion protection. The urgency level is assessed as high based on the inherent hazard of valve seal failure, the number of critical devices within the affected area, and the existing medium-risk status of the area in S1, and is then transmitted to the subsequent task allocation module.

[0046] In one embodiment of the present invention, step S3 includes the following steps: A risk propagation mathematical model is constructed, and the edge weights of the model are dynamically adjusted based on the wind flow direction in the dynamic risk situation semantic model to calculate the risk spread weight of each task. Extract the signal strength index of the region corresponding to each task, and convert the signal strength index into the probability of communication feasibility through a mapping function; The risk contagion weight is weighted by the communication feasibility probability to obtain the immediate risk utility score for each task.

[0047] First, a real-time task pool is obtained, consisting of periodic tasks generated by a pre-defined inspection plan and sudden tasks triggered by compound abnormal event entities. The real-time task pool is a dynamic dataset containing all currently pending inspection tasks, including planned periodic tasks and sudden tasks triggered by abnormal events. Pre-defined inspection plans are stored in a central server database in calendar or timetable format, defining inspection points, inspection frequency, and inspection content. The system's task scheduling service generates periodic tasks on schedule, for example, using a crontab or Quartz-based scheduler, and pushes them into a first-in-first-out (FIFO) queue as part of the task pool. Simultaneously, the task scheduling service monitors generated compound abnormal event entities in real-time through a message subscription mechanism. Upon receiving a new compound abnormal event entity, the service immediately parses it and automatically generates a sudden task with the highest priority based on its content. Sudden tasks and periodic tasks are placed together in a dynamically updated priority queue, sorted by task priority and generation timestamp, forming the real-time task pool.

[0048] Secondly, a risk propagation mathematical model is established, combining wind direction and equipment correlation from the dynamic risk situation semantic model to calculate the risk contagion weight of each task in the real-time task pool at the current moment. The risk propagation mathematical model is based on graph theory and is a mathematical tool used to simulate the dynamic propagation of risk within the oil depot across different equipment and areas due to environmental factors. This model is constructed as a directed weighted graph. , where the set of nodes The edge set represents all equipment and functional area entities within the defined oil depot. Represents the physical connection or spatial proximity between entities. Edge weight. It consists of two parts: static weights This indicates the degree of physical connection; for example, a directly connected pipe has a weight of 1, while a neighboring area has a weight of 0.5; and dynamic weights. Its value is adjusted in real time based on the real-time wind direction obtained in S1. Specifically, if the current wind direction is from node... Pointing to node Then the edge Dynamic weights It will increase proportionally with wind speed, while the opposite side The dynamic weights are reduced accordingly. For each task in the real-time task pool, the system determines the graph node corresponding to its target location. .

[0049] Risk contagion weight The calculation is based on this node In current mathematical models of risk propagation, risk centrality is a quantitative indicator ranging from 0 to 1, used to assess the risk correlation and potential propagation impact of a task's location. This centrality comprehensively considers the propagation from high-risk event source nodes to nodes through a weighted graph. The cumulative risk value, and from the node Starting from this point, the number and criticality of nodes affected by the potential downstream propagation of risk are calculated. This calculation is performed on a directed weighted graph using an improved PageRank algorithm or a heatmap diffusion algorithm, taking the urgency rating of the sudden task as the initial risk injection value, and finally obtaining the risk propagation weight for each task node. Its value is normalized to between 0 and 1.

[0050] Finally, signal strength indices for the corresponding regions of each task are extracted from the dynamic risk situation semantic model. The communication feasibility probability of each task is calculated, and the immediate risk utility score of each task is obtained by combining the risk contagion weight. For each task in the real-time task pool, the constructed dynamic risk situation semantic model is queried using its target geographic coordinates to obtain the real-time wireless communication signal strength index S at that location. This index is expressed in the form of Received Signal Strength Indicator (RSSI) in dBm. Subsequently, the signal strength index S is converted into a communication feasibility probability through a preset nonlinear mapping function. This probability is a value between 0 and 1, calculated based on the signal strength index. The nonlinear mapping function takes the form of the Sigmoid function:

[0051] in, It is a natural constant; The center point parameter is set to -85dBm, which represents the minimum signal strength threshold for maintaining stable backhaul. This is the sensitivity coefficient, with a numerical range set from 0.1 to 0.5, and its physical unit is [missing value]. Used to eliminate exponential terms The dimensions of the exponent ensure the mathematical rigor of exponential operations.

[0052] Finally, to prevent high-risk areas from being ignored by the system due to poor communication signals, a linear weighted summation method is used to calculate the instantaneous risk utility score. The formula is as follows:

[0053] in, As a risk contagion weight, Probability of communication feasibility; and These are the risk weight coefficient and the communication weight coefficient, respectively, and satisfy the following conditions: In this embodiment, to adhere to the principle of safety first, the following settings are made: , This score takes into account both the importance of task risk and communication feasibility, while ensuring that high-risk tasks receive a high execution priority even in weak communication environments.

[0054] For example, the generated composite abnormal event entity storage tank is used. valve Seal failure. Upon receiving this high-urgency-rating entity, the task scheduling service immediately generates an emergency task, such as TE01, with the task content being: [Regarding the valve...] High-precision imaging and spectral analysis were performed on the target location, which was a valve. The precise coordinates. Simultaneously, a pre-set inspection plan generates periodic tasks, such as TP01, whose task content is: to inspect the storage tanks. Perform routine temperature and liquid level checks, targeting the storage tank. The coordinates. These two tasks together constitute the current real-time task pool. Next, the risk contagion weights of these two tasks are calculated. For TE01, since it is directly associated with a high-urgency event and the current wind direction is 240°, the risk propagation mathematical model calculates the risk from the valve. The risk of propagation downwind is extremely high, therefore a risk contagion weight is assigned to it. It is 0.95.

[0055] For TP01, its target location is the storage tank. The risk contagion weight is as follows: Located upwind and currently without anomalies. The calculation result is 0.1. Subsequently, signal strength indices for the two task locations are extracted from the dynamic risk situation semantic model. Assume a valve... Due to its proximity to a large metal tank, the signal is blocked, and its signal strength index S is -88 dBm. (Storage tank) The location is open, and the signal strength specification S is -70 dBm. and The parameters are used to calculate the communication feasibility probability. For TE01, its communication feasibility probability is... For TP01, the probability of its communication feasibility is: .

[0056] Ultimately, based on and The weights are used to calculate the immediate risk utility scores for the two tasks. TE01's score is... The score for TP01 is... .

[0057] As can be seen from the comparison, although the communication conditions at the location of TE01 are poor, thanks to the dominant role of the risk weight in the weighted summation model, its final immediate risk utility score of 0.7712 is still significantly higher than that of the regular periodic task of 0.3559. This mechanism ensures that the system will prioritize scheduling robots to perform tasks in high-risk areas, and then rely on the local storage backup strategy in step S6 to solve the problem of weak network transmission, thereby eliminating the safety hazard of missing high-risk areas due to poor signal in the original logic.

[0058] In one embodiment of the present invention, step S4 includes the following steps: The system collects real-time data on the robot's remaining battery percentage, instantaneous power consumption rate, drive motor load status, and the online health status of onboard sensors to quantify hardware performance capabilities. Based on the current coordinates of the inspection robot, the dynamic risk situation semantic model is queried, and the available bandwidth of the current environment is predicted by combining historical network throughput data. The hardware execution capability indicators and the predicted available bandwidth are structured and encapsulated to generate a robot elastic resource vector.

[0059] First, the inspection robot's onboard management system collects real-time data on the remaining battery percentage, instantaneous power consumption rate, load status of each drive motor, and the online health status of each onboard sensor. The onboard management system is an embedded computer system that communicates with the robot's various subsystems via the vehicle's local area network. The remaining battery percentage is collected by accessing the robot's battery management system (BMS). The instantaneous power consumption rate refers to the rate at which the inspection robot consumes battery energy per unit time under current operating conditions; combined with the remaining battery percentage, the remaining driving time can be dynamically estimated. The BMS collects data every [period]. The remaining battery percentage is reported every second. The BMS reporting interval is specified below. The time interval is typically set to 1 to 5 seconds. Instantaneous power consumption rate refers to the rate at which the battery power of the inspection robot decreases per unit time, reflecting its current power consumption level and expected remaining operating time. The onboard management system maintains a time-series window, recording the most recent... Each power consumption reading is used to derive the current instantaneous power consumption rate through linear regression or differential calculation. The unit is percentage per minute.

[0060] The load status of each drive motor is acquired by reading the real-time current values ​​reported by each motor controller. To achieve this, the vehicle management system compares this current value with the motor's rated current. and locked rotor current Compare to determine the load status The load status of the drive motor is divided into four levels: idle, normal, high load, and overload. It is an assessment of the current stress level of the robot's moving parts to determine whether its mobility is limited.

[0061] Health status of each vehicle-mounted detection sensor Monitoring is performed via a heartbeat mechanism. This status provides a real-time assessment of the robot's perception system's functionality. Each onboard sensor operates at a fixed interval. Send a status packet to the vehicle management system if the preset timeout period has elapsed. If no status packet is received from a sensor, its status is marked as offline. Simultaneously, the status packet contains the sensor's internal self-test code; if the self-test code is abnormal, its status is marked as faulty. In this embodiment, the sensor heartbeat cycle... The timeout duration can be set to 1 second. This can be set to 3 seconds. Ultimately, the states of all sensors are aggregated into a comprehensive hardware health status. For example, it can be represented by a binary vector, where each bit corresponds to the online and normal state of a sensor.

[0062] Secondly, based on the current coordinates of the inspection robot, the wireless communication coverage density corresponding to the dynamic risk situation semantic model is analyzed to extract the predicted available bandwidth value for the current operating environment. The inspection robot obtains its own high-precision geographic coordinates in real time through its onboard RTK-GNSS positioning module and reports these coordinates to the vehicle management system. The vehicle management system uses these coordinates as a query index to request access to the constructed dynamic risk situation semantic model. In addition to storing static signal strength distribution maps, the dynamic risk situation semantic model dynamically records historical network congestion status and actual throughput data for specific areas, fed back by other fixed or mobile terminals. The signal strength index S of the robot's current location and its surrounding area is extracted and combined with historical throughput statistics. A pre-trained machine learning model, such as a gradient boosting decision tree or a small neural network, is used to predict the available bandwidth value for the current location. The available bandwidth prediction is an estimate of the wireless network bandwidth available for data transmission at the robot's current location, measured in Mbps. The model takes signal strength, historical average bandwidth, and current network load as input features and outputs a predicted bandwidth value in Mbps.

[0063] Finally, the remaining battery percentage, instantaneous power consumption rate, hardware health status, and predicted available bandwidth are structured to generate a robot elastic resource vector reflecting the robot's execution capabilities. The vehicle management system then uses the various indicators collected and predicted in the preceding steps, namely the remaining battery percentage... Instantaneous power consumption rate Overall hardware health status and the predicted available bandwidth This is integrated into a fixed-dimensional numerical vector. This vector is the robot's elastic resource vector. This vector is a multi-dimensional vector that structurally encapsulates the robot's real-time status in terms of power, hardware health, and communication capabilities, comprehensively representing its immediate ability to perform tasks. The robot's elastic resource vector is encapsulated in JSON format and periodically sent to the central task allocation module via a wireless communication link, serving as dynamic boundary constraints for subsequent task allocation decisions.

[0064] For example, suppose an inspection robot, Robot_01, is preparing to receive a task from the real-time task pool. First, Robot_01's onboard management system begins collecting its own resource status. By accessing the BMS, it obtains 12 battery readings over the past 60 seconds, linearly decreasing from 85.5% to 85.0%. The instantaneous power consumption rate is then calculated. The rate was 0.5% / min. Simultaneously, the real-time current of the robot's four drive motors was read to be between 30% and 50% of their rated values. Therefore, the load status of each drive motor was determined. The system is deemed normal. It also confirmed that it received heartbeat packets from all onboard sensors within the past 3 seconds, including the infrared thermal imager, gas detector, and vision camera, and that all self-check codes were normal. Therefore, the overall hardware health status is determined. Assuming all components are online and functioning normally, they can be represented by the vector [1, 1, 1, ...]. Next, Robot_01 reports its current coordinates as (121.495, 31.228, 1.5) m. The vehicle management system uses these coordinates to query the dynamic risk situation semantic model. The model returns the signal strength index S at this location as -75 dBm, and based on historical data analysis, predicts the available bandwidth at this location. The speed is 15 Mbps. Finally, the onboard management system performs structured processing on this data to generate a robot elastic resource vector. The vector is encapsulated as a JSON object. This vector is then sent to the central task allocation module, indicating that Robot_01's current power consumption is stable, its hardware is in good condition, its communication bandwidth is sufficient, and it has a good ability to perform various tasks.

[0065] In one embodiment of the present invention, step S5 includes the following steps: An optimization function is constructed with the goal of increasing the total weight score of tasks completed per unit time, and the robot's elastic resource vector is transformed into dynamic boundary constraints of power, hardware health and communication capabilities. By employing a multi-objective heuristic algorithm, an execution scheme is optimized that includes the sequence of movement path points, sensor activation periods, and data sampling density, while satisfying the access rules for the explosion-proof zones of the oil depot. The execution plan is encoded as a task dispatch instruction flow that controls the physical behavior of the robot.

[0066] First, an optimization function is constructed with the objective of increasing the total weight score of tasks completed per unit time. The immediate risk utility score is used as the core weight, and the robot's elastic resource vector is used as the dynamic boundary constraint. The optimization function is a mathematical expression used to quantify the merits of different task allocation schemes, aiming to find the scheme that maximizes this function value. The total weight score of tasks completed per unit time is the core indicator of the optimization function, reflecting the total value that can be obtained by the robot performing tasks within a given time. It is derived by weighted summation of the immediate risk utility scores of each task. The goal of the optimization function is to maximize the value within a preset time window. Within this context, it refers to the sum of weighted, immediate risk utility scores for all tasks performed by one or more inspection robots. It should be noted that... The total task duration is set based on the robot's battery capacity and workload, typically ranging from 3 to 8 hours. The instantaneous risk-utility score for each task is... This serves as the core weight coefficient in the optimization function for this task. The optimization function is expressed as: maximizing... ,in, It represents the total number of tasks awaiting completion in the real-time task pool. It is the first The immediate risk utility score of each task. It is a binary variable representing the task. Is it in The internal process is completed by robots.

[0067] Dynamic boundary constraints are derived from the robot's elastic resource vector. . Specifically, Instantaneous power consumption rate This is used to limit the total distance and working time of the robot to complete tasks, ensuring that it can complete the specified task set and return to the charging dock before the battery runs out. The load status of each drive motor and the online health status of each onboard detection sensor further limit the types and intensity of tasks the robot can undertake. For example, if the infrared thermal imager shows a malfunction, tasks involving infrared temperature measurement cannot be assigned to the robot. Available bandwidth prediction value The allocation of tasks requiring large amounts of data is constrained to ensure that communication quality meets task requirements. These resource parameters are transformed into a series of inequality constraints, limiting the feasible region of the optimization algorithm when searching for the optimal execution solution.

[0068] Secondly, a multi-objective heuristic algorithm is invoked to optimize the execution scheme, including the sequence of movement path points, sensor activation periods, and data sampling density, while satisfying the access rules of the oil depot's explosion-proof zone. The multi-objective heuristic algorithm is a computational method that can find approximate optimal solutions by simulating natural or physical processes when simultaneously optimizing multiple conflicting objectives. In this embodiment, an improved non-dominated sorting genetic algorithm II or a multi-objective particle swarm optimization algorithm can be used. This algorithm takes the optimization function as the primary objective and minimizes total energy consumption and total path length as secondary objectives. For example, if the NSGA-II algorithm is used, its population size is typically set to 50-100, the number of iterations is 100-200 generations, the crossover probability is 0.8, and the mutation probability is 0.1. The algorithm's input includes: a real-time task pool containing the immediate risk utility scores of each task, a robot elastic resource vector, and a constructed dynamic risk situation semantic model.

[0069] During optimization, the algorithm strictly considers the access rules for explosion-proof zones in oil depots. These rules are set according to national standards for different explosion-proof areas, specifying restrictions on entry for equipment and personnel with specific explosion-proof levels. Robots are prohibited from entering areas exceeding their explosion-proof level limits. If the dynamic risk situation semantic model detects that an area has been temporarily upgraded to a higher explosion-proof level, then the path planning of robots that do not meet that level must avoid that area. The algorithm iteratively generates potential inspection schemes, which are sequences consisting of a series of movement path points, sensor activation commands, and data sampling densities, and evaluates the objective function value and constraint satisfaction of each scheme. For path planning, the algorithm utilizes a high-precision 3D topology map and dynamic safety boundary distribution, combined with the A* or RRT* algorithm, to generate the shortest or optimal safe path from the robot's current position to the target task point. The output of the optimal execution scheme is a JSON or XML data structure containing: a complete sequence of movement path points for one or more robots, the precise activation time period for each sensor, and the corresponding data sampling density within each time period, such as the visual camera frame rate, gas sensor sampling frequency, and infrared image resolution.

[0070] Finally, the optimal execution plan is encoded into a task dispatch command stream that controls the robot's physical behavior. The task dispatch command stream is a set of encoded commands that the robot can directly execute, used to drive the robot's navigation, perception, and data transmission behaviors. The encoding process is implemented through an instruction generation module. This module receives the optimal execution plan and parses it into a serialized instruction set that the robot's underlying control system can directly recognize and execute. For example, each 3D coordinate pair in the movement path point sequence is converted into a navigation instruction, such as moving to X, Y, Z; sensor activation period instructions are converted into sensor control instructions, such as turning on the gas detector with a sampling frequency of 1 Hz and starting the infrared thermal imager with an image resolution of 640x480; data sampling density is converted into configuration instructions for the data processing unit. The task dispatch command stream uses a custom binary protocol or a message format based on the Robot Operating System to ensure the efficiency and reliability of instruction transmission. The command stream also includes a unique identifier for the task, the expected completion time, and error handling codes to support subsequent task execution monitoring and feedback.

[0071] For example, using the previously calculated TE01 and TP01, as well as the robot elastic resource vector of Robot_01... An optimization function is constructed to maximize the sum of immediate risk utility scores for task completion, with the weight of TE01 updated to 0.7712 and the weight of TP01 updated to 0.3559. Dynamic boundary constraints include: Robot_01's current remaining battery power is 85%, the instantaneous power consumption rate is 0.5% / min, the maximum supported cruising range and operating time; all onboard sensors are online and functioning normally, therefore all sensor-related tasks can be executed; the predicted available bandwidth is 15 Mbps. The NSGA-II algorithm is used for optimization. The algorithm starts from Robot_01's current position (121.490, 31.225, 0.5) m and considers including TE01 and TP01 in the execution plan. When generating the path scheme, the optimizer strictly queries the updated dynamic risk situation semantic model.

[0072] Due to the target area valve of TE01 The area has been temporarily upgraded to Zone 0. The algorithm ensures that the path planned for Robot_01 will not cross this restricted area and selects the safest and most efficient arrival path. Meanwhile, for TE01, due to its high risk of valve seal failure requiring detailed analysis, the optimizer will plan sensor activation periods, including immediately activating the high-resolution vision camera for video recording upon arrival at the target point, and simultaneously activating the gas detector, with a data sampling density set to 20 frames per second of high-frequency images and gas readings every 0.1 seconds. For TP01, due to its lower priority and available communication bandwidth, the algorithm plans sensor activation periods of regular photography and temperature readings, with a data sampling density set to 1 frame per second of images and a temperature reading every 1 second. Finally, the optimization algorithm outputs the optimal execution plan. The system then encodes this optimal execution plan into a task dispatch instruction stream. These instructions will be packaged and sent to Robot_01 via a wireless communication link, guiding it to execute the corresponding inspection tasks.

[0073] In one embodiment of the present invention, step S6 includes the following steps: Monitor the real-time communication quality during the robot's execution process. When the communication quality is lower than the preset bandwidth threshold, adjust the data transmission granularity and feed it back to step S3 to correct the communication feasibility probability of subsequent tasks. Identify the updated equipment status and environmental parameters during the inspection task execution process, and feed them back to step S1 to correct the attribute status and safety boundary distribution in the dynamic risk situation semantic model.

[0074] First, the inspection robot is driven to perform on-site operations according to the task allocation command stream, and the task execution progress and sensor measurement data are uploaded in real time. The inspection robot receives the task allocation command stream through its onboard communication module. The command stream is parsed by the robot's task actuator and converted into a sequence of low-level control commands. The robot's navigation module uses onboard LiDAR scanning and an inertial measurement unit (IMU) for high-precision positioning and map matching, driving the robot to move along the movement path point sequence defined in the command stream. During the robot's movement, its perception system drives onboard gas detectors, infrared thermal imagers, and vision cameras to collect data based on the sensor activation periods and data sampling densities defined in the command stream. For example, if the command stream requires high-resolution video recording in a specific area, the vision camera will acquire images at a high frame rate and high resolution. The onboard edge computing unit performs real-time preprocessing on the acquired raw sensor data, such as image compression, gas reading calibration, and anomaly detection and early warning. The task executor monitors the robot's current position, the number of completed waypoints, sensor status, and battery level in real time. It then uploads this task progress information, along with pre-processed sensor data, to the central control platform via a wireless communication link using the MQTT protocol. Task progress refers to the physical state and logical progression of the inspection robot in completing its task sequence; sensor data refers to the raw or semi-raw data collected in real time by the robot's various onboard sensors and pre-processed. The data upload frequency is set to 1 to 5 times per second, depending on the criticality of the task.

[0075] Secondly, the system monitors real-time communication quality during execution. When real-time communication quality falls below a preset bandwidth threshold, it automatically adjusts data transmission granularity and local storage backup strategies, and feeds this status back to correct the feasibility rating of subsequent tasks. The robot's onboard communication module continuously monitors the real-time communication quality metrics of the currently connected wireless network, including RSSI, SNR, and actual UDP / TCP data throughput. The onboard management system compares the actual data throughput with a series of preset bandwidth thresholds. These bandwidth thresholds are preset and used to determine the minimum data transmission rate required to meet specific data transmission needs.

[0076] If the actual data throughput is continuous This is below the bandwidth threshold set for high-definition video streaming. If the system determines that the real-time communication quality has deteriorated, an automatic adjustment mechanism is triggered: the data transmission granularity is reduced. For example, the data stream frame rate of the vision camera automatically decreases from 25 fps to 5 fps, and the image resolution automatically decreases from 1920x1080 to 640x480. It should be understood that data transmission granularity refers to the level of detail or volume of data during transmission, such as the resolution, frame rate, and sampling frequency of images or videos. Simultaneously, a local storage backup strategy is activated, meaning that all non-urgent data that can be delayed is no longer transmitted immediately but is cached on the robot's built-in industrial-grade solid-state drive.

[0077] When communication quality recovers to above When the robot moves to an area with better communication quality, the cached data will be automatically uploaded in batches. Furthermore, the vehicle management system encapsulates the current robot location, actual data throughput, and communication quality degradation status into feedback messages, for example, indicating that the effective bandwidth has decreased to... The feedback is transmitted to the central task allocation module via a wireless communication link. Upon receiving this feedback, the central task allocation module uses the decreased communication quality as dynamic input to recalculate or adjust the communication feasibility probability of the robot's current position and its surrounding area. This allows for the correction of the immediate risk utility score for subsequent tasks involving the region.

[0078] Finally, the updated equipment status and environmental parameters during the inspection process are fed back in real time to correct the dynamic risk situation semantic model and achieve dynamic rolling optimization of the task allocation strategy. During the on-site operation of the inspection robot, its onboard sensors, through real-time perception and edge analysis, may detect and identify new equipment status or environmental parameter changes. Updated equipment status refers to changes in the current operating status of the equipment detected and confirmed by the inspection robot through sensors during the inspection process; environmental parameters refer to physical or chemical quantities related to the oil depot environment, such as temperature, humidity, and gas concentration, which are acquired in real time by the inspection robot through sensors. For example, the robot's infrared thermal imager might detect during inspection that the surface temperature at a pipe connection is higher than the baseline value recorded by the static semantic baseline. Temperatures in Celsius, or a gas detector detecting that the methane concentration in a specific area is consistently higher than [a certain value]. These updated device statuses and environmental parameters are identified by the robot's field perception module.

[0079] The field perception module transmits this structured update information to the dynamic risk situation semantic model update service in real time via a wireless communication link. Upon receiving this information, the update service immediately corrects the dynamic risk situation semantic model. For example, it updates the state attribute of the corresponding pipeline entity in the dynamic risk situation semantic model to local overheating, and triggers a recalculation of the gas diffusion model based on newly detected methane concentration, which may lead to a redefinition of the dynamic safety boundary distribution or an adjustment of the risk level of the relevant area. This corrected dynamic risk situation semantic model will immediately perform task priority assessment and task allocation decisions, thus forming continuous closed-loop management and achieving dynamic rolling optimization of the task allocation strategy. Dynamic rolling optimization refers to the process of enabling the task allocation strategy to continuously adapt to changes in the oil depot environment and equipment status through real-time feedback and model correction.

[0080] For example, using the task dispatch instruction stream generated by Robot_01, it instructs Robot_01 to execute TE01, that is, to the storage tank. valve Perform high-precision imaging and spectral analysis. Robot_01 begins to move towards the valve according to the command flow. The robot moves, and during this process, it uploads its current coordinates and battery percentage, among other task progress data, in real time. When Robot_01 moves to the distance valve... At a distance of approximately 10 meters, the vehicle's communication module detected a sharp decline in real-time communication quality, with the actual data throughput measured at 2.5 Mbps for three consecutive tests, far below the bandwidth threshold. At this point, Robot_01's automatic adjustment mechanism is triggered. The data transmission granularity is reduced; the frame rate of the high-definition vision camera video stream, originally set at 20 fps, is automatically adjusted to 5 fps, and the image resolution is reduced to 640x480. Simultaneously, a local cache task is created, and all non-critical, routine area inspection image data is marked as delayed for upload and temporarily stored in the robot's built-in SSD. Robot_01 also reports the current communication quality degradation and its current position to the central task allocation module.

[0081] The central task allocation module's instruction service updates the communication feasibility probability of the area. This causes subsequent tasks involving the area to receive a negative correction with a lower communication weight when calculating their immediate risk utility scores. Through this mechanism, the system can automatically reduce the priority of high-bandwidth-demand tasks in that area or prompt the scheduling of robots with stronger communication capabilities. Robot_01 then continued executing TE01 as its infrared thermal imager scanned the valve. At that time, the surface temperature of the valve was detected to be 88℃, far exceeding the recorded normal operating temperature of 58℃, with a temperature difference of 30℃, exceeding [the threshold]. The temperature was 10 degrees Celsius. Meanwhile, the valve was detected by a high-precision gas detector. The nearby methane concentration was 75 ppm, consistently higher than [previous value]. The level was 50 ppm. The inspection robot's on-site sensing module determined that this was a valve... Localized overheating and gas leakage caused by seal failure. Structured update information is sent in real-time to the dynamic risk situation semantic model update service.

[0082] Upon receiving this information, the service immediately corrects the dynamic risk situation semantic model: [regarding the valve]. The status attribute is updated from potential leakage to seal failure, and the gas diffusion model is recalculated more accurately based on the new gas leak points and concentration data. The dynamic safety boundary distribution is dynamically adjusted, potentially allowing the explosion-proof level of the area to be temporarily upgraded to a safer management level. This revised dynamic risk situation semantic model will be used in the next task allocation decision, thereby achieving dynamic rolling optimization of the task allocation strategy and ensuring that robot scheduling can continuously respond to the latest risk dynamics of the oil depot.

[0083] See appendix Figure 2 The present invention also proposes an IoT-based task allocation system for an oil depot inspection robot, comprising the following modules: The composite anomaly diagnosis module is used to integrate multimodal heterogeneous data from inspection robots and fixed sensors, and identify composite anomaly entities through logical reasoning. The task risk utility calculation module is used to calculate the real-time risk utility score of each task in the real-time task pool by combining the dynamic risk situation semantic model. The robot resource vectorization module is used to collect the operating status and current location communication characteristics of the inspection robot and generate the robot's elastic resource vector. The task allocation instruction generation module is used to solve an optimization function to generate a task allocation instruction flow by using the real-time risk utility score as the weight and the robot's elastic resource vector as the constraint. The closed-loop strategy adaptive module drives the robot to operate according to the task allocation instruction flow, and adaptively corrects the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

[0084] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A task allocation method for an IoT-based oil depot inspection robot, characterized in that, Includes the following steps: S1. Based on the topological information and real-time environmental parameters of the target oil depot, construct a dynamic risk situation semantic model; S2. Integrates multimodal heterogeneous data from inspection robots and fixed sensors to identify complex abnormal event entities through logical reasoning; S3. Combine the dynamic risk situation semantic model to calculate the instantaneous risk utility score of each task in the real-time task pool. S4. Collect the operating status and current location communication characteristics of the inspection robot to generate a robot elastic resource vector; S5. Using the immediate risk utility score as the weight and the robot's elastic resource vector as the constraint, solve the optimization function to generate the task allocation instruction flow; S6. Drive the robot to operate according to the task allocation instruction flow, and adaptively correct the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

2. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S1 includes the following steps: Obtain a 3D topological map of the target oil depot, define the explosion-proof level of each functional area and the physical spatial coordinates of the equipment through attribute annotation, and form a static semantic basis; Real-time monitoring data from fixed IoT sensors is retrieved to obtain real-time environmental parameters, including ambient temperature and humidity, wind direction, and concentration of characteristic gases. By integrating real-time environmental parameters into a static semantic foundation, and updating the physical facility status and mapping dynamic security boundaries, a dynamic risk situation semantic model is generated.

3. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S2 includes the following steps: Perform spatiotemporal correlation analysis on multimodal heterogeneous data to extract sensor data clusters that are within a preset time window and whose spatial distance is less than a preset distance threshold; By using a chain-based reasoning model combined with a pre-set fault feature library, the causal relationship between different sensor physical indicators in the data cluster is determined. Based on the judgment results, the root cause failure mode is identified, and the scope of risk impact and urgency level rating are determined by combining the dynamic risk situation semantic model, generating composite abnormal event entities.

4. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S3 includes the following steps: A risk propagation mathematical model is constructed, and the edge weights of the model are dynamically adjusted based on the wind flow direction in the dynamic risk situation semantic model to calculate the risk spread weight of each task. Extract the signal strength index of the region corresponding to each task, and convert the signal strength index into the probability of communication feasibility through a mapping function; The risk contagion weight is weighted by the communication feasibility probability to obtain the immediate risk utility score for each task.

5. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S4 includes the following steps: The system collects real-time data on the robot's remaining battery percentage, instantaneous power consumption rate, drive motor load status, and the online health status of onboard sensors to quantify hardware performance capabilities. Based on the current coordinates of the inspection robot, the dynamic risk situation semantic model is queried, and the available bandwidth of the current environment is predicted by combining historical network throughput data. The hardware execution capability indicators and the predicted available bandwidth are structured and encapsulated to generate a robot elastic resource vector.

6. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S5 includes the following steps: An optimization function is constructed with the goal of increasing the total weight score of tasks completed per unit time, and the robot's elastic resource vector is transformed into dynamic boundary constraints of power, hardware health and communication capabilities. By employing a multi-objective heuristic algorithm, an execution scheme is optimized that includes the sequence of movement path points, sensor activation periods, and data sampling density, while satisfying the access rules for the explosion-proof zones of the oil depot. The execution plan is encoded as a task dispatch instruction flow that controls the physical behavior of the robot.

7. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, Step S6 includes the following steps: Monitor the real-time communication quality during the robot's execution process. When the communication quality is lower than the preset bandwidth threshold, adjust the data transmission granularity and feed it back to step S3 to correct the communication feasibility probability of subsequent tasks. Identify the updated equipment status and environmental parameters during the inspection task execution process, and feed them back to step S1 to correct the attribute status and safety boundary distribution in the dynamic risk situation semantic model.

8. The task allocation method for an IoT-based oil depot inspection robot according to claim 1, characterized in that, The specific steps for adjusting the granularity of data transmission include: Reduce the data stream frame rate and image resolution of the vision camera, and activate the local storage backup strategy to cache non-urgent data to the robot's built-in storage medium, and upload it in batches after the communication quality is restored.

9. A task allocation method for an IoT-based oil depot inspection robot according to claim 3, characterized in that, The specific steps for determining causal relationships include: The data clusters are input into a Bayesian network model based on a directed acyclic graph, and the posterior probability of each potential fault event is calculated. When the posterior probability exceeds a preset decision threshold, it is determined that there is a causal relationship between the sensor physical indicators.

10. A task allocation system for an oil depot inspection robot based on the Internet of Things, characterized in that, Includes the following modules: The risk situation semantic modeling module constructs a dynamic risk situation semantic model based on the topological information and real-time environmental parameters of the target oil depot. The composite anomaly diagnosis module is used to integrate multimodal heterogeneous data from inspection robots and fixed sensors, and identify composite anomaly entities through logical reasoning. The task risk utility calculation module is used to calculate the real-time risk utility score of each task in the real-time task pool by combining the dynamic risk situation semantic model. The robot resource vectorization module is used to collect the operating status and current location communication characteristics of the inspection robot and generate the robot's elastic resource vector. The task allocation instruction generation module is used to solve an optimization function to generate a task allocation instruction flow by using the real-time risk utility score as the weight and the robot's elastic resource vector as the constraint. The closed-loop strategy adaptive module drives the robot to operate according to the task allocation instruction flow, and adaptively corrects the dynamic risk situation semantic model and the real-time risk utility score based on the execution feedback.

Images