A method and system for optimizing the scheduling of unmanned aerial vehicle (UAV) highway inspection tasks

Abstract

This invention relates to the fields of intelligent transportation and UAV scheduling technology, and provides a method and system for optimizing UAV highway inspection task scheduling. The method includes the following steps: real-time acquisition of dynamic status information and environmental constraint information of all available UAVs, forming a set of UAV resource states and a set of environmental constraints; based on the initial inspection task set, constructing a spatiotemporal correlation graph of the highway network, and inferring and generating derivative predictive tasks to obtain a task pool to be scheduled; establishing and solving a mixed-integer programming model based on the UAV resource state set, the environmental constraint set, and the task pool to be scheduled, and assigning an inspection task plan to each UAV; real-time monitoring of the UAV's execution of the inspection task plan, generating predictive secondary tasks and execution deviation, and determining whether rescheduling is triggered. This invention achieves a fundamental transformation in highway UAV inspection from passive response to proactive prediction, and from static planning to dynamic adaptation.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation and unmanned aerial vehicle (UAV) scheduling technology, and in particular relates to a method and system for optimizing the scheduling of UAV highway inspection tasks. Background Technology

[0002] Highway inspection is a crucial link in ensuring road network safety and smooth traffic. With the development of drones and intelligent sensing technologies, existing technologies have proposed scheduling methods that integrate multi-source data to generate inspection tasks and allocate tasks to drone fleets based on multi-objective optimization models, aiming to improve inspection efficiency and emergency response capabilities.

[0003] However, existing optimization scheduling methods are mostly based on a single, static decision-making framework. This means that at the beginning of a single scheduling cycle, a one-time optimization solution is performed based on a fixed set of tasks and the UAV's state at the current moment, generating a task plan for the entire cycle. This method fails to fully consider the dynamic changes during task execution: on the one hand, the UAV's state (such as location and battery level) continuously changes while performing tasks, and new events may occur at any time; on the other hand, the state of discovered events may deteriorate or generate new secondary risks, and the priority and handling requirements of the original tasks may change. Static scheduling schemes often exhibit rigidity and lag when dealing with such a highly dynamic environment, failing to achieve optimal resource utilization and event handling effectiveness.

[0004] Therefore, there is an urgent need for an inspection task optimization method that can incorporate real-time feedback and perform dynamic rescheduling, so that the UAV scheduling process can continuously adapt to changes in the environment and task status. Summary of the Invention

[0005] The purpose of this invention is to provide a method for optimizing the scheduling of unmanned aerial vehicle (UAV) highway inspection tasks, thereby solving the aforementioned technical problems.

[0006] This invention is implemented as follows: a method for optimizing the scheduling of unmanned aerial vehicle (UAV) highway inspection tasks, comprising the following steps:

[0007] Real-time acquisition of multi-source data containing highway event information to obtain the raw event information stream;

[0008] The original event information stream is fused and deduplicated, and the dynamic priority score of each event is quantified to generate an initial inspection task set.

[0009] Real-time acquisition of dynamic status information and environmental constraint information of all available drones, forming a set of drone resource status and a set of environmental constraints;

[0010] Based on the initial inspection task set, a spatiotemporal correlation graph of the highway network is constructed, and by traversing the spatiotemporal correlation graph, derivative predictive tasks are generated to obtain a task pool to be scheduled.

[0011] Based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, a mixed integer programming model is established and solved to generate a scheduling scheme to allocate inspection task plans to each UAV.

[0012] The process of the UAV executing the inspection task plan is monitored in real time, predictive secondary tasks and execution deviation are generated, and it is determined whether to trigger rescheduling.

[0013] Furthermore, the steps of fusing and deduplicating the original event information stream, quantifying the dynamic priority score of each event, and generating an initial inspection task set specifically include:

[0014] Based on the time, geographical location, and type characteristics of the events, cross-source comparison and correlation analysis are performed on the original event information stream, and the results are merged and deduplicated to obtain a standardized event list;

[0015] For each event in the standardized event list, obtain the predicted traffic impact and the probability of escalation of the situation in the future period after the event occurs, and generate a dynamic priority score.

[0016] Based on the type, geographical location, dynamic priority score, and preset handling procedures of each event, a structured inspection task entry is generated for each event, resulting in an initial set of inspection tasks.

[0017] Furthermore, based on the initial inspection task set, a spatiotemporal correlation graph of the highway network is constructed, and by traversing the spatiotemporal correlation graph, derivative predictive tasks are generated to obtain a task pool to be scheduled. This process specifically includes:

[0018] Using the spatiotemporal attributes of all events in the initial inspection task set as nodes, a spatiotemporal relationship graph of the highway network is constructed.

[0019] Define the weights of the edges between event nodes in the spatiotemporal correlation graph;

[0020] Based on graph traversal algorithms, derivative predictive tasks with predictive properties are generated by reasoning according to the weights of the edges.

[0021] All the derived predictive tasks generated by reasoning are merged with the initial inspection task set to form a task pool to be scheduled.

[0022] Furthermore, the weight of the edge is defined as: within a preset time difference and spatial distance range, the probability of the occurrence or continuation of an event causing or aggravating the occurrence of another event.

[0023] Furthermore, the mixed integer programming model aims to minimize the total global task response time, the number of unfinished high-priority tasks, and the total energy consumption of the UAV.

[0024] Furthermore, the method for generating the predictive secondary task is as follows:

[0025] During the execution of the inspection mission plan by the UAV, real-time inspection data transmitted back by the UAV is received in real time; based on the real-time inspection data, spatiotemporal correlation diagram and current mission, the causal impact probability of the event corresponding to the current mission being upgraded to an event with a higher dynamic priority score is predicted; if the causal impact probability exceeds the preset warning threshold, a predictive secondary mission with a higher priority is generated.

[0026] Furthermore, the method for generating the execution deviation is as follows:

[0027] Continuously compare the execution status of the drone with the target status in the inspection task plan; based on the comparison results, determine the execution deviation.

[0028] Then, based on the dynamic priority score and execution deviation of the predictive secondary task, it is determined whether to trigger rescheduling.

[0029] Another objective of this invention is to provide a UAV highway inspection task scheduling optimization system for implementing the aforementioned UAV highway inspection task scheduling optimization method, comprising:

[0030] The event information acquisition module is used to acquire multi-source data containing highway event information in real time to obtain the raw event information stream;

[0031] The inspection task generation module is used to fuse and deduplicate the original event information stream, quantify the dynamic priority score of each event, and generate an initial set of inspection tasks.

[0032] The status and constraint acquisition module is used to acquire the dynamic status information and environmental constraint information of all available UAVs in real time, forming a set of UAV resource status and a set of environmental constraints.

[0033] The task-to-be-scheduled determination module is used to construct a spatiotemporal correlation graph of the highway network based on the initial inspection task set, and to generate derived predictive tasks by traversing the spatiotemporal correlation graph to obtain a task-to-be-scheduled pool.

[0034] The scheduling scheme generation module is used to establish and solve a mixed integer programming model based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, and generate a scheduling scheme to allocate inspection task plans to each UAV.

[0035] The rescheduling trigger module is used to monitor the process of the UAV executing the inspection task plan in real time, generate predictive secondary tasks and execution deviation, and determine whether to trigger rescheduling.

[0036] This invention provides a method for optimizing the scheduling of UAV highway inspection tasks, achieving a fundamental shift in highway UAV inspection from passive response to proactive prediction, and from static planning to dynamic adaptation. This invention can integrate multi-source event information in real time, proactively generate derived prediction tasks based on spatiotemporal correlation graphs, and optimize the allocation of UAV resources under multiple constraints using a mixed-integer programming model. Furthermore, by introducing real-time monitoring and a closed-loop feedback mechanism, this invention can dynamically trigger rescheduling based on changes in the on-site situation and predictive secondary tasks during task execution, thereby significantly improving the initiative, timeliness, and resource utilization efficiency of highway emergency inspections. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the unmanned aerial vehicle (UAV) highway inspection task scheduling optimization method provided in an embodiment of the present invention.

[0038] Figure 2 This is a flowchart illustrating step S200 in the UAV highway inspection task scheduling optimization method provided in an embodiment of the present invention.

[0039] Figure 3 This is a flowchart illustrating step S400 in the UAV highway inspection task scheduling optimization method provided in this embodiment of the invention.

[0040] Figure 4 This is a schematic diagram of the structure of the UAV highway inspection task scheduling optimization system provided in an embodiment of the present invention. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0042] like Figure 1 As shown, in one embodiment of the present invention, a method for optimizing the scheduling of UAV highway inspection tasks is provided, comprising the following steps:

[0043] S100: Real-time acquisition of multi-source data containing highway event information to obtain the raw event information stream;

[0044] S200. The original event information stream is fused and deduplicated, the dynamic priority score of each event is quantified, and an initial inspection task set is generated.

[0045] S300: Real-time acquisition of dynamic status information and environmental constraint information of all available UAVs, forming a set of UAV resource status and a set of environmental constraints;

[0046] S400. Based on the initial inspection task set, construct a spatiotemporal correlation graph of the highway network, and generate derivative predictive tasks by traversing the spatiotemporal correlation graph to obtain a task pool to be scheduled.

[0047] S500. Based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, establish and solve a mixed integer programming model to generate a scheduling scheme for allocating inspection task plans to each UAV.

[0048] S600 monitors the process of the UAV executing the inspection task plan in real time, generates predictive secondary tasks and execution deviation, and determines whether to trigger rescheduling.

[0049] In practical applications, in step S100, data from multiple sources, such as the drone AI recognition system, roadside fixed sensing devices (such as cameras and radars), and manual reporting platforms, are acquired in real time. The multi-source data is then processed for time synchronization, geographic coordinate alignment, and format standardization to form a unified description of the original event information stream.

[0050] It should be noted that an event refers to an abnormal situation or specific state that occurs in and around the highway network and its surrounding environment and needs to be perceived and may require handling. In practical applications, the types of events include traffic accidents (such as rear-end collisions, rollovers, collisions with guardrails, etc.), traffic anomalies (such as abnormal congestion, illegal parking, driving against traffic, pedestrians or animals running into traffic, etc.), infrastructure failures (such as severe potholes on the road surface, large obstacles such as fallen cargo, damaged signs, landslides, etc.), environmental safety (such as sudden drops in visibility due to severe weather, road icing, fires, smoke, etc.), and planned inspection targets (such as including planned inspection points such as regular inspections of bridges and tunnels as special preset events in the task pool).

[0051] like Figure 2 As shown, in a preferred embodiment of the present invention, the initial inspection task set includes event location, type, dynamic priority score, and expected handling method; wherein, the expected handling method includes preliminary handling action and expected handling target state; the step of fusing and deduplicating the original event information stream, quantifying the dynamic priority score of each event, and generating the initial inspection task set, i.e., step S200, specifically includes:

[0052] S210. Based on the time, geographical location and type characteristics of the event, perform cross-source comparison and correlation analysis on the original event information stream, and fuse and deduplicate to obtain a standardized event list;

[0053] Specifically, repeated alarm information describing the same target or scenario within a preset spatiotemporal threshold is fused, and conflicting information is arbitrated according to the data source confidence rules, outputting a standardized event list that is free of redundancy and complementary in information.

[0054] S220. For each event in the standardized event list, obtain the predicted traffic impact value and the probability value of escalation of the situation in the future period after the event occurs, and generate a dynamic priority score.

[0055] Specifically, by combining the type, geographical location, and preliminary severity rating of each event, and by calling the real-time traffic flow prediction model and the traffic accident prediction model, the predicted traffic impact and the probability of escalation of the event in the future are obtained. After normalizing the multi-dimensional features such as event type (different event types can be assigned different values ​​based on a predefined event type library), preliminary severity rating (the severity of the event can be initially rated manually), traffic impact prediction, and probability of escalation, a linear weighted fusion is performed (the fusion weight of each feature can be determined by training historical data or manually based on the actual situation). This allows for the calculation of a dynamic priority score that comprehensively reflects the urgency, importance, and timeliness requirements of the event.

[0056] The real-time traffic flow prediction model employs a hybrid architecture combining a spatiotemporal graph convolutional network (ST-GCN) and a gated recurrent unit (GRU). This model constructs a spatiotemporal graph based on the highway network topology, with nodes representing road segments or detector locations and edges indicating connections between segments. Input data consists of multi-dimensional features such as traffic flow, average speed, and occupancy from historical multi-step time series. The spatiotemporal graph convolutional layer captures the spatial dependencies of the road network, and the GRU layer models the dynamic evolution patterns over time. The model uses a multi-task learning framework to simultaneously predict traffic conditions (flow and speed) at three time scales: 15, 30, and 60 minutes in advance. Training utilizes historical traffic detector data and weather calendar data, employing mean squared error and mean absolute percentage error as loss functions. A sliding time window is used to generate training samples, and end-to-end training is performed using backpropagation and the Adam optimizer. An early-stop strategy is employed to prevent overfitting. This real-time traffic flow prediction model can simultaneously capture both the static topological constraints of the road network and the dynamic propagation patterns of traffic flow, achieving high-precision short-term traffic flow prediction.

[0057] Furthermore, a traffic accident prediction model, based on an integrated framework of deep survival analysis and gradient boosting trees (XGBoost), can output the probability of a traffic accident occurring on a specific road segment within a given future time period. This model first discretizes continuous time into multiple time intervals and uses a deep neural network (DNN) to learn the accident risk feature representation for each time interval. The input to the traffic accident prediction model is a high-dimensional feature vector, including real-time traffic flow parameters (such as flow rate mutation rate and speed variation coefficient), road segment static attributes (number of lanes, curvature), environmental factors (weather, lighting), historical accident frequency, and time series patterns. Feature interaction and selection are performed through fully connected layers and an attention mechanism to output the conditional risk probability for each time interval. During training, historical accident records are used as positive samples (event occurrence time is known), a negative log-likelihood loss function is employed, and a time-related regularization term is introduced to handle data censoring. Finally, by integrating the risk outputs from multiple time intervals, the traffic accident prediction model obtains the accident probability curve for the next 1-60 minutes, providing a quantitative risk basis for dynamic priority calculation.

[0058] S230. Based on the type, geographical location, dynamic priority score, and preset handling procedures of each event, generate a structured inspection task item for each event to obtain an initial inspection task set.

[0059] The aforementioned inspection task items must include at least the event location, event type, dynamic priority score, and expected handling method. All inspection task items together constitute the initial inspection task set for subsequent steps. In addition, the expected handling method includes the initial handling actions required to be performed by the drone upon arrival (such as hovering surveillance, taking photos for evidence, and issuing verbal warnings to guide traffic) and the expected handling target state that is desired for the event (such as restoring traffic flow and confirming and eliminating potential hazards).

[0060] In this embodiment of the invention, by constructing a standardized multi-source data real-time fusion and dynamic task generation process, the shortcomings of information silos, duplicate alarms, and static priority assessment in traditional highway inspections are effectively solved. This embodiment of the invention introduces a dynamic priority quantification mechanism that combines a real-time traffic flow prediction model and a traffic accident prediction model, enabling task sequencing to accurately reflect the urgency, potential for hazard development, and handling value of events in a specific spatiotemporal environment. This transforms previously scattered and isolated event information into an initial set of inspection tasks that can directly drive optimized scheduling.

[0061] In a preferred embodiment of the present invention, step S300 specifically includes:

[0062] First, the dynamic status information of each available UAV in the system is acquired in real time through the UAV ground control station, airport management system, and airborne data link. This dynamic status information includes at least: the UAV's real-time geographic coordinates and altitude, remaining battery power or fuel, the current working status and health of each airborne mission device (such as cameras, loudspeakers, and robotic arms), and the load of its currently assigned but not yet completed ordered task sequence. Simultaneously, the meteorological service system and airspace management system are acquired to obtain detailed real-time weather data (such as wind speed, precipitation, and visibility) and short-term forecasts covering the current location of all UAVs and the task areas involved in the initial inspection task set, as well as temporary or permanent airspace restriction information (such as no-fly zones, restricted flight altitudes, and busy airspace periods), forming environmental constraint information.

[0063] Then, the acquired raw dynamic status information is cleaned and formatted. Based on the UAV model and equipment configuration, its remaining battery power is converted into the estimated remaining flight time under typical operating modes. The capability matching degree for performing various tasks (such as reconnaissance, surveillance, and delivery) in the initial inspection task set is determined based on the status of its onboard equipment, yielding capability matching degree information. This capability matching degree information is spatially correlated and timestamped with environmental constraint information, generating a dynamically updated resource status entry for each UAV. All resource status entries are aggregated to form the UAV resource status set. Simultaneously, the spatiotemporally correlated environmental constraint information is structured to form an environmental constraint set. These two sets together serve as a complete snapshot of the availability of scheduling resources and external constraints in the current and short-term future, providing real-time and accurate input for subsequent optimized scheduling.

[0064] like Figure 3 As shown, in a preferred embodiment of the present invention, the step of constructing a spatiotemporal correlation graph of the highway network based on the initial inspection task set, and generating derived predictive tasks by traversing the spatiotemporal correlation graph to obtain a task pool to be scheduled, specifically step S400, includes:

[0065] S410. Using the spatiotemporal attributes (such as geographical location, occurrence or expected occurrence time) of all events in the initial inspection task set as nodes, construct a spatiotemporal correlation graph of the highway network. This spatiotemporal correlation graph not only includes the physical topology of the network, but also extends the causal relationships in the time dimension.

[0066] S420. Define the weights of the edges between event nodes in the spatiotemporal correlation graph;

[0067] Specifically, the weight of an edge is defined as the probability that, within a preset time difference and spatial distance, the occurrence or continuation of an event (source node) triggers or exacerbates the occurrence of another event (target node). In practical applications, the weights of edges between event nodes in the spatiotemporal relationship graph can be calculated based on statistical data of historical events, using the following formula:

[0068] ;

[0069] ;

[0070] In the formula, Let be the weight of the edge between event node i and event node j; Let be the causal influence probability between event node i and event node j; This represents the number of times event j occurs within a preset time window after event i occurs in historical data. This represents the total number of times event i occurred in the historical data. The type similarity between event i and event j can be determined by the cosine similarity of the event type features; The spatiotemporal constraint function is determined as follows: If the distance between event node i and event node j is not greater than the preset maximum spatial influence distance (e.g., 5 kilometers), and the time difference between the occurrences of event node i and event node j is not greater than the preset maximum time influence window (e.g., 30 minutes), then... ;otherwise, .

[0071] S430. Based on the graph traversal algorithm, a derivative predictive task with predictive properties is generated by reasoning based on the weight of the edges.

[0072] Specifically, a graph traversal algorithm (such as depth- or breadth-based search) is used to explore along edges with significant causal probability, starting from each event node in the initial inspection task set. For potential event points that are reachable within the set spatiotemporal influence radius but are not yet in the initial inspection task set, the probability of them being triggered is determined through reasoning. If the probability exceeds a preset threshold, a derivative predictive task with predictive properties is generated, whose dynamic priority score can be partially inherited from the source event and adjusted with reference to the prediction probability.

[0073] S440. Merge all the derived predictive tasks generated by reasoning with the initial inspection task set to form a task pool to be scheduled; this task pool is more comprehensive, including events that have occurred and high-probability potential events.

[0074] In this embodiment of the invention, by introducing spatiotemporal correlation graphs and causal probability reasoning, it is possible to proactively identify and quantify the potential derivative and transmission risks between different events, thereby generating derivative predictive tasks in advance. This significantly improves the forward-looking scheduling capability, enabling drone resources to be deployed to high-risk areas in advance, effectively preventing the occurrence of secondary accidents or curbing the chain reaction of the situation.

[0075] In a preferred embodiment of the present invention, the hybrid integer partitioning model aims to minimize the total global task response time, the number of unfinished high-priority tasks, and the total energy consumption of the UAV.

[0076] In practical applications, the inspection task plan includes an ordered task sequence and waypoint and time schedule; step S500 specifically includes:

[0077] First, the aforementioned set of UAV resource states and environmental constraints, along with the pool of tasks to be scheduled, are used as input data. Based on this input data, key parameters are set for the mixed-integer programming model. These key parameters include: the starting position and endurance of each UAV, the position and expected processing time window of each task, the capability matching matrix between UAVs and tasks, the spatial safety interval requirements between tasks, and the task weight coefficient reflecting the dynamic priority score. Next, based on the aforementioned key parameters, a mixed-integer programming model is constructed with the core objectives of minimizing the total global task response time, minimizing the number of uncompleted high-priority tasks, and minimizing the total UAV energy consumption (in terms of total flight time). The expression for the mixed-integer programming model is as follows;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] In the formula, F1 is the total response time index of the global mission; F2 is the index of the number of high-priority missions that have not been completed; and F3 is the total energy consumption index of the UAV (based on total flight time). , , The target weight coefficients, the sum of which is 1, can be determined according to actual needs; generally, The value ranges from 0.2 to 0.3. The value ranges from 0.4 to 0.6. The value ranges from 0.2 to 0.3; S T S represents the task pool to be scheduled, where T is the total number of tasks in the task pool;H S represents a subset of high-priority tasks in the task pool to be scheduled (including tasks with dynamic priority scores higher than a preset threshold), where H is the total number of high-priority tasks in the subset; U Let U be the set of drone resource states, and U be the total number of drones in the set of drone resource states. The start time of task i; The time of the first report or discovery of task i; t max This is the maximum allowable response time, used to normalize the overall response time metric for the global task; z i Indicates whether task i is completed; a value of 1 indicates completion, and a value of 0 indicates failure. ij Let be the flight distance from mission i to mission j; This indicates whether drone k travels from task i to task j; the value is 1 if yes and 0 otherwise. This represents the average flight speed of the drone k; Let k be the maximum flight time of the drone, used to normalize the total flight time of the drone.

[0083] This mixed-integer programming model uses the drone's endurance, the matching requirements of onboard equipment and mission type, and airspace safety interval rules as hard constraints. Its decision variables include the drone-mission allocation relationship, mission execution order, and precise start and end times.

[0084] Then, a mathematical optimization solver (such as CPLEX, Gurobi) is used to solve the mixed integer programming model constructed above. By weighing multiple competing objectives, the mathematical optimization solver outputs an optimal or near-optimal scheduling scheme under the current constraints. This scheduling scheme explicitly specifies the ordered sequence of tasks assigned to each scheduled UAV during the planning period, as well as the precise start time, end time, and critical waypoints of the flight path for each task.

[0085] Finally, the obtained scheduling scheme is transformed into a structured instruction set that can be recognized by the UAV flight control system and operators, namely, an inspection task plan containing an ordered sequence of tasks and precise waypoint and time schedules. This inspection task plan serves as the benchmark execution scheme for the current scheduling cycle, and is prepared to be issued to the corresponding UAVs for execution and synchronized to the monitoring system for subsequent process tracking.

[0086] In a preferred embodiment of the present invention, the method for generating the predictive secondary task is as follows:

[0087] During the execution of the inspection mission plan by the UAV, real-time inspection data transmitted back by the UAV is received in real time; based on the real-time inspection data, spatiotemporal correlation diagram and current mission, the causal impact probability of the event corresponding to the current mission being upgraded to an event with a higher dynamic priority score is predicted; if the causal impact probability exceeds the preset warning threshold, a predictive secondary mission with a higher priority is generated.

[0088] In a preferred embodiment of the present invention, the method for generating the execution deviation is as follows:

[0089] Continuously compare the execution status of the drone with the target status in the inspection task plan; based on the comparison results, determine the execution deviation.

[0090] Then, based on the dynamic priority score and execution deviation of the predictive secondary task, it is determined whether to trigger rescheduling.

[0091] In practical applications, step S600 specifically includes:

[0092] First, as the UAV executes its mission according to the aforementioned ordered task sequence and precise waypoint and time plan, it receives real-time inspection data, including on-site video, sensor data, and mission status. This real-time inspection data is combined with a spatiotemporal correlation diagram and the expected handling method for the current task to analyze the actual evolution trend of the discovered event. The probability of the event escalating into a higher-level event or triggering derivative secondary events in the subsequent period is predicted. If this probability exceeds a preset warning threshold, a higher-priority predictive secondary task is generated and inserted into the real-time task pool. Simultaneously, the actual flight trajectory of the UAV, mission start / end time, and on-site event status are continuously compared with the expected waypoints, time nodes, and expected handling target status defined in the inspection task plan. Based on these comparison results, an execution deviation rate reflecting progress delays, geographical deviations, and differences in handling effectiveness is calculated. Specifically, the execution deviation is calculated as follows: The time deviation is determined based on the difference between the actual time and the planned time for key nodes of the inspection task (such as arrival at the task point, start of operation, and end of operation); the spatial deviation is determined based on the difference in Euclidean distance between the actual geographical location of the UAV at the key waypoint or the inspection task execution location and the planned location; the degree of conformity between the event handling result and the expected handling target state defined in the expected handling method is evaluated based on the real-time data transmitted back by the UAV (such as images and sensor readings) to determine the task effect deviation (e.g., 1 for conformity, zero for non-conformity); after normalizing the time deviation, spatial deviation, and task effect deviation, they are weighted and summed according to their respective preset weight coefficients to obtain the execution deviation.

[0093] Then, the dynamic priority score and execution deviation of the generated predictive secondary tasks are monitored in real time; the rescheduling process is triggered immediately when any of the following conditions are met:

[0094] 1. The dynamic priority score of the predictive secondary mission exceeds the dynamic priority score of the mission currently being performed by the affected UAV;

[0095] 2. The execution deviation of any drone or mission exceeds the preset tolerance threshold.

[0096] Once rescheduling is triggered, the system immediately takes the latest set of UAV resource states, environmental constraints, and a new pool of tasks to be scheduled (which integrates the real-time pool of tasks to be scheduled, incomplete original tasks, and newly triggered predictive secondary tasks) as input, calls the mixed integer programming model again to solve the problem, generates a new inspection task plan starting from the current moment, and immediately updates and distributes it to the relevant UAVs, thus forming a dynamic closed-loop feedback control loop.

[0097] In this embodiment of the invention, by constructing a dynamic closed-loop feedback control loop embedded in the task execution process, the fundamental defects of traditional static scheduling schemes—rigidity and inability to adapt to real-time changes—are solved. This embodiment proactively predicts event evolution and generates predictive secondary tasks, enabling forward-looking risk management. Through real-time monitoring of execution deviations and an intelligent judgment triggering mechanism, the system can promptly correct deviations between planned and actual inspection tasks. Finally, through closed-loop adaptive rescheduling, continuous optimal or near-optimal allocation of UAV resources is achieved in dynamically changing environments, greatly improving robustness and timeliness in responding to emergencies and handling complex chain events.

[0098] like Figure 4 As shown, in another embodiment of the present invention, a UAV highway inspection task scheduling optimization system is also provided to implement the above-mentioned UAV highway inspection task scheduling optimization method, which includes:

[0099] Event information acquisition module 10 is used to acquire multi-source data containing highway event information in real time to obtain the raw event information stream;

[0100] The inspection task generation module 20 is used to fuse and deduplicate the original event information stream, quantify the dynamic priority score of each event, and generate an initial inspection task set.

[0101] The state and constraint acquisition module 30 is used to acquire the dynamic state information and environmental constraint information of all available UAVs in real time, forming a set of UAV resource states and a set of environmental constraints.

[0102] The task-to-be-scheduled determination module 40 is used to construct a spatiotemporal correlation graph of the highway network based on the initial inspection task set, and to generate derived predictive tasks by traversing the spatiotemporal correlation graph to obtain a task-to-be-scheduled pool.

[0103] The scheduling scheme generation module 50 is used to establish and solve a mixed integer programming model based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, and generate a scheduling scheme to allocate inspection task plans to each UAV.

[0104] The rescheduling trigger module 60 is used to monitor the process of the UAV executing the inspection task plan in real time, generate predictive secondary tasks and execution deviation, and determine whether to trigger rescheduling.

[0105] It should be noted that each of the above modules can be implemented as a computer program, which can run on a computer device. The computer device's memory can store the computer program consisting of each module, enabling the processor to execute each step of the above method.

[0106] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0107] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods.

[0108] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for optimizing the scheduling of unmanned aerial vehicle (UAV) highway inspection tasks, characterized in that, Includes the following steps: Real-time acquisition of multi-source data containing highway event information to obtain the raw event information stream; The original event information stream is fused and deduplicated, and the dynamic priority score of each event is quantified to generate an initial inspection task set. Real-time acquisition of dynamic status information and environmental constraint information of all available drones, forming a set of drone resource status and a set of environmental constraints; Based on the initial inspection task set, a spatiotemporal correlation graph of the highway network is constructed, and by traversing the spatiotemporal correlation graph, derivative predictive tasks are generated to obtain a task pool to be scheduled. Based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, a mixed integer programming model is established and solved to generate a scheduling scheme to allocate inspection task plans to each UAV. The process of the UAV executing the inspection task plan is monitored in real time, predictive secondary tasks and execution deviation are generated, and it is determined whether rescheduling is triggered. Based on the initial inspection task set, the process of constructing a spatiotemporal correlation graph of the highway network, and generating derived predictive tasks by traversing the spatiotemporal correlation graph to obtain a task pool to be scheduled, specifically includes: Using the spatiotemporal attributes of all events in the initial inspection task set as nodes, a spatiotemporal relationship graph of the highway network is constructed. Define the weights of the edges between event nodes in the spatiotemporal correlation graph; Based on graph traversal algorithms, derivative predictive tasks with predictive properties are generated by reasoning according to the weights of the edges. All the derived predictive tasks generated by reasoning are merged with the initial inspection task set to form a task pool to be scheduled. The weight of the edge is defined as: within a preset time difference and spatial distance range, the probability of the causal influence of the occurrence or continuation of an event causing or aggravating the occurrence of another event. The method for generating the predictive secondary task is as follows: During the execution of the inspection mission plan by the UAV, real-time inspection data transmitted back by the UAV is received in real time; based on the real-time inspection data, spatiotemporal correlation diagram and current mission, the causal impact probability of the event corresponding to the current mission being upgraded to an event with a higher dynamic priority score is predicted; if the causal impact probability exceeds the preset warning threshold, a predictive secondary mission with a higher priority is generated.

2. The method for optimizing the scheduling of UAV highway inspection tasks according to claim 1, characterized in that, The steps of fusing and deduplicating the original event information stream, quantifying the dynamic priority score of each event, and generating an initial inspection task set specifically include: Based on the time, geographical location, and type characteristics of the events, cross-source comparison and correlation analysis are performed on the original event information stream, and the results are merged and deduplicated to obtain a standardized event list; For each event in the standardized event list, obtain the predicted traffic impact and the probability of escalation of the situation in the future period after the event occurs, and generate a dynamic priority score. Based on the type, geographical location, dynamic priority score, and preset handling procedures of each event, a structured inspection task entry is generated for each event, resulting in an initial set of inspection tasks.

3. The method for optimizing the scheduling of UAV highway inspection tasks according to claim 1, characterized in that, The hybrid integer programming model aims to minimize the total global task response time, the number of unfinished high-priority tasks, and the total energy consumption of the UAV.

4. The method for optimizing the scheduling of UAV highway inspection tasks according to claim 1, characterized in that, The method for generating the execution deviation is as follows: Continuously compare the execution status of the drone with the target status in the inspection task plan; based on the comparison results, determine the execution deviation. Then, based on the dynamic priority score and execution deviation of the predictive secondary task, it is determined whether to trigger rescheduling.

5. A UAV highway inspection task scheduling optimization system, used to implement the UAV highway inspection task scheduling optimization method according to any one of claims 1-4, characterized in that, include: The event information acquisition module is used to acquire multi-source data containing highway event information in real time to obtain the raw event information stream; The inspection task generation module is used to fuse and deduplicate the original event information stream, quantify the dynamic priority score of each event, and generate an initial set of inspection tasks. The status and constraint acquisition module is used to acquire the dynamic status information and environmental constraint information of all available UAVs in real time, forming a set of UAV resource status and a set of environmental constraints. The task-to-be-scheduled determination module is used to construct a spatiotemporal correlation graph of the highway network based on the initial inspection task set, and to generate derived predictive tasks by traversing the spatiotemporal correlation graph to obtain a task-to-be-scheduled pool. The scheduling scheme generation module is used to establish and solve a mixed integer programming model based on the set of UAV resource states, the set of environmental constraints, and the pool of tasks to be scheduled, and generate a scheduling scheme to allocate inspection task plans to each UAV. The rescheduling trigger module is used to monitor the process of the UAV executing the inspection task plan in real time, generate predictive secondary tasks and execution deviation, and determine whether to trigger rescheduling.

Images