Air-ground vehicle networking double-time-scale scheduling method based on situation awareness

Abstract

The application discloses a kind of based on situation awareness's air-ground vehicle networking double time scale arrangement method. First, the air-ground vehicle networking edge cloud collaborative execution environment consisting of roadside edge server, unmanned aerial vehicle air edge node and cloud node is constructed, and business is modeled as the directed acyclic graph DAG workflow with task dependency relationship. Secondly, the local observation information such as link communication quality, service processing capacity, fault danger exposure and energy state of each computing node is collected, and network level situation state for unified scheduling decision is generated by time smoothing, neighborhood fusion and uncertainty calibration. Then, based on the network level situation state and execution trajectory, checkpoint configuration, backup resource reservation, recovery mode preference and deadline tail risk budget are updated on slow time scale, and ready task main execution node selection, backup node selection, transmission path selection, communication resource allocation, computing resource allocation and recovery action selection are executed on fast time scale;When detecting node failure, task deadline cannot be met or communication contact relationship fails, trigger backup switching, checkpoint playback, path re-routing or task migration and other online recovery processes. Finally, according to task actual execution result, update physical task queue, virtual risk queue, unmanned aerial vehicle residual energy state and execution trajectory set, form the closed-loop arrangement mechanism of protection adaptive and online recovery collaborative coupling.

Description

Technical Field

[0001] This invention relates to the fields of air-ground cooperative vehicle-to-everything (V2X) networks, intelligent scheduling of edge-cloud continuums, UAV-assisted edge computing, and elastic control of complex task flows. Specifically, it relates to a dual-timescale orchestration method for air-ground V2X networks based on situational awareness. This invention is applicable to air-ground cooperative business processing environments comprised of roadside infrastructure, aerial UAV platforms, and cloud computing centers, and particularly relates to directed acyclic graphs (DAGs) for scenarios such as cooperative perception, assisted driving, road event detection, dynamic inspection, emergency rescue, and risk warning. Integrated orchestration and control of task mapping, resource allocation, protection configuration and online recovery of task flow. Background Technology

[0002] With the rapid development of intelligent transportation systems, vehicle-road cooperative systems, and drone-assisted perception technologies, air-ground cooperative vehicle networks, composed of roadside units, vehicle-mounted terminals, drone platforms, and cloud computing centers, are gradually becoming important infrastructure supporting complex real-time business processing. In this type of system, roadside infrastructure can provide relatively stable edge computing and communication access capabilities, drone platforms can quickly supplement local area perception, forwarding, and computing resources due to their flexible and mobile deployment advantages, and cloud computing centers can undertake global management, long-term optimization, and cross-regional collaborative tasks. Thus, an edge-cloud continuum business processing architecture with multi-layered heterogeneous resource collaboration characteristics is formed.

[0003] In application scenarios such as collaborative perception, assisted driving, road event detection, dynamic inspection, emergency rescue, and risk warning, the tasks to be processed are usually not single, independent tasks, but rather complex task flows composed of multiple sub-tasks with sequential dependencies. Such task flows can often be abstracted as a directed acyclic graph (DAG). In this context, there are clear data dependencies, execution order, and latency constraints between different task nodes. Since some tasks require the output of preceding tasks to continue execution, and different tasks have different requirements for computing power, communication bandwidth, completion time limits, and reliability assurance, how to achieve efficient, reliable, and flexible orchestration of complex task flows in an air-ground cooperative vehicle-to-everything (V2X) environment has become a crucial research problem in the field of edge intelligent scheduling. Existing technologies for V2X, mobile edge computing, and UAV-assisted edge computing mostly focus on issues such as task offloading, resource allocation, link selection, path planning, or local scheduling, which can improve system throughput, reduce average service latency, or increase resource utilization to some extent. However, most existing solutions are based on a single time scale for decision-making, typically performing real-time scheduling based only on the local state within the current time slot or short time window, lacking a comprehensive consideration of the system's long-term operating state, risk accumulation process, and protection strategy adjustment mechanisms. When network topology, link quality, node load, and UAV remaining energy continuously change, relying solely on a single time scale scheduling method makes it difficult to balance short-term response capability with long-term stability.

[0004] Furthermore, in the air-ground collaborative vehicle-to-everything (V2X) edge-cloud continuum environment, the execution of complex task flows faces significant uncertainty and vulnerability due to the continuous changes in UAV flight positions, the susceptibility of wireless links to obstruction and interference, the significant heterogeneity of edge node service capabilities, and the potential for cross-node and cross-regional failures caused by common causes. For example, some tasks may fail to complete as planned due to the failure of the main execution node, link interruption, unmet time limits, or loss of contact relationships, leading to task migration, duplicate transmission, or even the failure of the entire task flow. Without protection configurations and online recovery mechanisms adapted to the system's operating state, it is difficult to guarantee the continuous execution capability of complex services in a dynamic environment. In addition, existing technical solutions for reliable scheduling, fault-tolerant orchestration, or workflow recovery often employ static backups, fixed checkpoint cycles, single-copy redundancy, or partial rescheduling to improve system robustness. While these methods can enhance task execution reliability to some extent, they typically have the following shortcomings:

[0005] 1) The lack of a unified global situational awareness mechanism makes it difficult to integrate multi-source information such as link status, service capabilities, fault risks, and energy status into a unified decision-making basis;

[0006] 2) Protection strategies are often fixed in advance or adjusted with coarse granularity, making it difficult to dynamically update checkpoints, backups, and risk budgets based on business execution trajectories and system state evolution results;

[0007] 3) The online recovery process is disconnected from the initial protection configuration, making it difficult to achieve effective synergy between protection costs and recovery benefits;

[0008] 4) For tasks with strict time limits and task dependencies Existing workflow methods struggle to simultaneously balance execution efficiency, survivability, recovery costs, and tail risk control.

[0009] Therefore, there is an urgent need to propose an orchestration method suitable for the air-ground cooperative vehicle-to-everything (V2X) edge-cloud continuum environment, capable of handling environments with dependencies between different vehicles. The task flow comprehensively utilizes link communication status, node processing capabilities, fault hazard exposure, and energy security information to construct a unified situational awareness mechanism. Based on this, it combines long-term protection strategy adjustments with short-term online orchestration control to achieve coordinated optimization of task mapping, resource allocation, protection configuration, and online recovery, thereby improving the execution reliability and overall scheduling performance of complex services in dynamic, heterogeneous, and uncertain environments. Summary of the Invention

[0010] The purpose of this invention is to provide a situational awareness-based dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks, addressing the following problems in the execution of complex task flows in the edge-cloud continuum environment of existing air-to-ground collaborative V2X networks: significant link state fluctuations, heterogeneous service capabilities of edge nodes, limited remaining energy of UAVs, and difficulty in uniformly handling cross-regional common-cause faults. Furthermore, most existing methods are based on local scheduling at a single timescale, making it difficult to balance long-term protection configurations with short-term online recovery, resulting in directed acyclic graphs with dependencies between different components. Task flow in dynamic environments suffers from problems such as high task completion latency, high recovery costs, insufficient survivability, and high risk of defaulting on deadlines.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] A situational awareness-based dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks, characterized by the following steps:

[0013] 1) Construct a collaborative execution environment for the air-ground vehicle-to-everything (V2X) network edge-cloud continuum. This collaborative execution environment includes roadside edge servers, aerial drone nodes, and a cloud computing center. The business to be processed is modeled as a directed acyclic graph with task dependencies. The workflow consists of tasks with different computational requirements and completion deadlines, and each task's dependent edge has corresponding data transmission requirements.

[0014] 2) Collect local observation information of each computing node and communication link. The local observation information includes at least link communication quality, node service capability, fault hazard exposure degree and energy status. Generate network-level situational status through time smoothing, neighborhood fusion and uncertainty calibration, so as to serve as a unified decision basis for subsequent task orchestration and recovery control.

[0015] 3) Based on the network-level situation and historical execution trajectory, update the protection strategy on a slow time scale. The protection strategy includes at least checkpoint configuration, backup resource reservation, recovery mode preference and tail risk budget of deadline, which are used to provide long-term protection parameters for task flow execution in subsequent time windows.

[0016] 4) Based on the network-level situational awareness of the current time slot and the updated protection strategy, perform primary execution node selection, backup node selection, transmission path selection, communication resource allocation, and computing resource allocation for ready tasks on a fast time scale to form the online orchestration result of the current time slot; when it is detected that the primary execution node of the task is unavailable, the remaining time limit of the task cannot be met, or the communication contact relationship is invalid, trigger an online recovery process including backup switching, checkpoint replay, path rerouting, and task migration.

[0017] 5) During mission execution and recovery, the computational energy consumption, communication energy consumption, and maneuvering energy consumption of UAV aerial nodes are considered simultaneously. When node capacity constraints, link capacity constraints, energy security constraints, or protection constraints are not met, resource projection and feasibility repair are performed on the orchestration results. Based on the actual mission execution results, the physical mission queue, survivability virtual queue, tail risk virtual queue, UAV remaining energy status, and execution trajectory set are updated, and the updated results are fed back to the protection strategy update process in the subsequent time window, thereby forming a dual-timescale closed-loop orchestration mechanism that coordinates long-term protection configuration and short-term online recovery.

[0018] In the above technical solution, the air-ground vehicle-to-everything (V2X) edge-cloud continuum collaborative execution environment constructed in step 1) further includes the following four parts:

[0019] (1) Roadside edge service module: Deployed on the roadside edge server, it provides low-latency computing, caching and communication access capabilities for vehicle terminals, roadside sensing devices and aerial drone nodes, and is responsible for undertaking some task node execution, task-dependent data forwarding and local resource coordination;

[0020] (2) Airborne UAV Coordination Module: Deployed on the UAV platform, it provides mobile perception, airborne forwarding and auxiliary computing capabilities when roadside infrastructure coverage is insufficient, event area is dynamically changing or local links are blocked, and collects information on the UAV's remaining energy, flight position, service load and link status.

[0021] (3) Cloud-based global control module: Deployed in the cloud computing center, used to perform cross-regional global policy maintenance, historical execution trajectory statistics, slow time scale protection parameter updates, and multi-regional collaborative scheduling management; (4) Workflow modeling and orchestration control module: Used to abstract the business to be processed into a directed acyclic graph. The workflow, combined with the computational requirements of task nodes, the data requirements of dependent edges, the node capability status, the link status, and the risk status, completes task mapping, resource allocation, protection configuration, and online recovery control.

[0022] The set of computing nodes in the air-to-ground vehicle network edge-cloud continuum is defined as follows:

[0023]

[0024] in, This represents the set of roadside edge servers. Represents a set of aerial drone nodes. This represents a cloud computing center; the communication connections between the nodes constitute a set of links. The pending business processes are modeled as a directed acyclic graph workflow. ,in, Represents a set of task nodes. Represents the set of task-dependent edges; for any task node Its computational requirements are defined as follows: For any dependent edge The amount of data to be transmitted is defined as Workflow release time is defined as The completion deadline is defined as .

[0025] Furthermore, in step 2), the network-level situational awareness is constructed as follows:

[0026] For any computing node Collect it in time slot The local observation vector is defined as:

[0027]

[0028] in, This represents the observation value of the communication quality of the link related to the node. This represents the observed value of the node's service processing capacity. This represents the observed value of packet transmission performance. Represents the observed value of the node's remaining energy state.

[0029] This represents the dangerous exposure observation value of the fault domain to which the node belongs.

[0030] To reduce the impact of transient disturbances on subsequent scheduling decisions, time smoothing is performed on the local observation vector to obtain the smoothed local situation parameters:

[0031]

[0032] in, This is the smoothing coefficient.

[0033] set up Represents a node In the time slot The set of communicating neighbors, then the node The fusion state quantity is represented as:

[0034]

[0035] in, Let be the neighborhood fusion weight, and satisfy:

[0036]

[0037] Based on the fused situational information of all nodes, construct the network-level situational status:

[0038]

[0039] in, This represents the calibrated estimate of the low-quantum link capability. This represents the calibrated estimate of low quantile service capacity. Represents the fault domain Risk rate estimate Indicates the total cost of the link. This represents an uncertainty indicator. Furthermore, the overall link cost is expressed as:

[0040]

[0041] in, , and These represent link latency cost, hazard cost, and maneuver cost, respectively.

[0042] This represents the corresponding weighting coefficient.

[0043] Furthermore, in step 3), the slow time-scale protection strategy is updated as follows:

[0044] In the window The protection strategy is defined as follows:

[0045]

[0046] in, Represents the set of protected dependent edges. Represents the periodic vector of checkpoints. Indicates tail risk budget, This represents the recovery mode preference vector. For any candidate dependency edge... and candidate checkpoint cycle Its protection cost is defined as:

[0047]

[0048] in, This indicates basic protection costs. Indicates the checkpoint cycle is Estimated protection overhead at that time Display window The above is a scene dispersion statistic extracted from the execution trajectory. Correspondingly, the protection benefit is defined as:

[0049]

[0050] in, This indicates the amount of improvement in recovery latency. This indicates the amount of improvement in tail risk. This represents the improvement in survivability. Further, the slow timescale scoring function is defined as:

[0051]

[0052] Based on the scoring results of each candidate dependency edge under different candidate checkpoint periods, the checkpoint period with the highest score is selected, and the set of protected dependency edges is determined under the protection budget constraint. Furthermore, the window-level tail risk budget is updated as follows:

[0053]

[0054] in, Represents the interval projection operator. Display window Tail risk statistics within.

[0055] Furthermore, in step 4), the fast timescale online orchestration and online recovery method is as follows:

[0056] For any ready task Define the main execution node selection variable. And backup node selection variables ,satisfy:

[0057]

[0058]

[0059] in, Represents a node Selected as a mission The main execution node, Represents a node Selected as a mission Backup nodes. For dependent edges. If its transmission path is Then its transmission delay is:

[0060]

[0061] in, Indicates link The allocated transmission rate. If the task... At the node If executed above, its computation delay is:

[0062]

[0063] in, Indicates assignment to a node Service speed. Task The arrival time of the precursor is expressed as:

[0064]

[0065] Therefore, the task The completion time is expressed as:

[0066]

[0067] in, Indicates queuing delay. This indicates the recovery of additional latency. Further, the end-to-end completion latency of the workflow is expressed as:

[0068]

[0069] in, Indicates the completion time of the virtual exit node. Defines the recovery trigger indicator. for:

[0070]

[0071] in, Indicates the availability of the primary execution node. Indicates the remaining time limit margin for the task. Indicates the feasibility of contact; when At that time, an online recovery process is triggered, including backup switching, checkpoint replay, path rerouting, and task migration. For candidate actions... Its scoring function is defined as follows:

[0072]

[0073] in, This indicates the predicted execution and recovery costs. This indicates an increase in tail risk.

[0074] The survivability gain is represented by minimizing the scoring function to determine the task orchestration result for the current time slot.

[0075] Furthermore, in step 5), the energy safety constraints, survivability constraints, and closed-loop feedback update methods are as follows:

[0076] The computing nodes in the system are divided into a set of fault domains. ,Task In the fault domain The deployment instruction quantity is defined as follows:

[0077]

[0078] Fault Domain In the prediction time domain The survival probability within is expressed as:

[0079]

[0080] Therefore, the task-level survivability metric is expressed as:

[0081]

[0082] The workflow is required to meet the following requirements:

[0083]

[0084] in, This represents the workflow survivability target threshold. For drone nodes... Its calculated energy consumption is expressed as:

[0085]

[0086] Its communication energy consumption is expressed as:

[0087]

[0088] The drone's remaining battery energy has been updated to:

[0089]

[0090] And satisfy:

[0091]

[0092] in, The energy safety threshold for drones.

[0093] 6) Joint optimization objectives and constraints:

[0094] The dual-time-scale orchestration method aims to minimize the long-term average completion delay of the workflow, and the objective function is expressed as:

[0095]

[0096] It also simultaneously satisfies constraints on computing capacity, bandwidth capacity, workflow survivability, deadline tail risk, and UAV energy safety.

[0097] 7) After each time slot or time window ends, the physical task queue, survivability virtual queue, tail risk virtual queue, UAV remaining energy status and execution trajectory set are updated according to the actual task execution results. The updated results are fed back to the protection strategy update process of the subsequent time window, thus forming a dual-timescale closed-loop orchestration mechanism that coordinates long-term protection configuration and short-term online recovery.

[0098] The inventive principle of this invention:

[0099] This invention addresses the complex edge-cloud continuum environment of air-to-ground vehicle-to-everything (V2X) networks. To address the challenges of link fluctuations, node heterogeneity, UAV energy constraints, and cross-regional common-cause failures encountered during task flow execution, a situational awareness-based dual-timescale orchestration mechanism is proposed. Its fundamental principles are: First, by fusing multi-source information such as link communication status, node service capabilities, fault hazard exposure, and remaining energy status, a network-level situational awareness reflecting the overall network operation is constructed, providing a unified and stable decision-making basis for task orchestration and recovery decisions. Second, at the slow timescale, checkpoint configuration, backup reservation, recovery preferences, and tail risk budget are adaptively updated based on historical execution trajectories to achieve dynamic adjustment of long-term protection configurations. Third, at the fast timescale, master node selection, backup node selection, path selection, resource allocation, and recovery action selection are performed for currently ready tasks to cope with time-slot-level network changes and task execution disturbances. Finally, by detecting abnormal situations such as node failures, link interruptions, unmet time limits, and insufficient energy, and triggering corresponding online recovery actions, a closed-loop collaboration between protection configuration and recovery execution is achieved. Thus, under the premise of ensuring system capacity, survivability, energy security, and risk constraints, a system can be implemented in an air-to-ground vehicle-to-everything (V2X) environment. Efficient, reliable, and flexible execution of task flows.

[0100] The beneficial effects of this invention are as follows:

[0101] This invention proposes for the first time a situational awareness-based dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks. This method can handle sequential dependencies in the edge-cloud continuum environment of air-to-ground cooperative V2X networks. The system responds to task flow processing requests and dynamically enhances the execution and recovery capabilities of complex services in heterogeneous environments through collaborative scheduling among roadside edge nodes, aerial drone nodes, and cloud computing centers. Under the premise of meeting constraints such as computing capacity, link bandwidth, workflow survivability, deadline tail risk, and drone energy safety, the system aims to improve the execution stability, reliability, and timeliness of complex task flows. It establishes a dual-timescale closed-loop collaborative relationship between "network-level situational awareness construction," "slow-timescale protection strategy updates," and "fast-timescale online orchestration and recovery control." Furthermore, it extends to scenarios with significant link fluctuations, heterogeneous node service capabilities, dynamically changing fault risks, and limited aerial node energy. It introduces fault domain modeling, survivability assessment, checkpoint configuration, backup reservation, and online recovery mechanisms to output a set of stable and feasible task orchestration and recovery schemes under the current network conditions, thereby achieving complex... Adaptive, reliable, and elastic execution of task flows in a dynamic air-ground collaborative environment. Attached Figure Description

[0102] Figure 1 This is a schematic diagram of the system architecture of the air-ground vehicle-to-everything (V2X) edge-cloud continuum collaborative execution environment of the present invention;

[0103] Figure 2 This is a schematic diagram of the dual-time-scale closed-loop orchestration framework of the present invention;

[0104] Figure 3 This is a schematic diagram illustrating the fault domain modeling and task survivability assessment of the present invention.

[0105] Figure 4 A schematic diagram illustrating the impact of different numbers of workflow instances on the performance of the method of the present invention and the comparative method;

[0106] Figure 5 A schematic diagram illustrating the impact of different workflow scales on the performance of the method of the present invention and the comparative method;

[0107] Figure 6 A schematic diagram illustrating the impact of different communication computation ratios on the performance of the method of the present invention and the comparative method;

[0108] Figure 7 This diagram illustrates the impact of different numbers of edge nodes on the performance of the method of this invention and the comparative method. Detailed Implementation

[0109] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0110] A situational awareness-based dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks, characterized by the following steps:

[0111] 1) Construct a collaborative execution environment for the air-ground vehicle-to-everything (V2X) network edge-cloud continuum. This collaborative execution environment includes roadside edge servers, aerial drone nodes, and a cloud computing center. The business to be processed is modeled as a directed acyclic graph with task dependencies. The workflow consists of tasks with different computational requirements and completion deadlines, and each task's dependent edge has corresponding data transmission requirements.

[0112] 2) Collect local observation information of each computing node and communication link. The local observation information includes at least link communication quality, node service capability, fault hazard exposure degree and energy status. Generate network-level situational status through time smoothing, neighborhood fusion and uncertainty calibration, so as to serve as a unified decision basis for subsequent task orchestration and recovery control.

[0113] 3) Based on the network-level situation and historical execution trajectory, update the protection strategy on a slow time scale. The protection strategy includes at least checkpoint configuration, backup resource reservation, recovery mode preference and tail risk budget of deadline, which are used to provide long-term protection parameters for task flow execution in subsequent time windows.

[0114] 4) Based on the network-level situational awareness of the current time slot and the updated protection strategy, perform primary execution node selection, backup node selection, transmission path selection, communication resource allocation, and computing resource allocation for ready tasks on a fast time scale to form the online orchestration result of the current time slot; when it is detected that the primary execution node of the task is unavailable, the remaining time limit of the task cannot be met, or the communication contact relationship is invalid, trigger an online recovery process including backup switching, checkpoint replay, path rerouting, and task migration.

[0115] 5) During mission execution and recovery, the computational energy consumption, communication energy consumption, and maneuvering energy consumption of UAV aerial nodes are considered simultaneously. When node capacity constraints, link capacity constraints, energy security constraints, or protection constraints are not met, resource projection and feasibility repair are performed on the orchestration results. Based on the actual mission execution results, the physical mission queue, survivability virtual queue, tail risk virtual queue, UAV remaining energy status, and execution trajectory set are updated, and the updated results are fed back to the protection strategy update process in the subsequent time window, thereby forming a dual-timescale closed-loop orchestration mechanism that coordinates long-term protection configuration and short-term online recovery.

[0116] In the above technical solution, the air-ground vehicle-to-everything (V2X) edge-cloud continuum collaborative execution environment constructed in step 1) further includes the following four parts:

[0117] (1) Roadside edge service module: Deployed on the roadside edge server, it provides low-latency computing, caching and communication access capabilities for vehicle terminals, roadside sensing devices and aerial drone nodes, and is responsible for undertaking some task node execution, task-dependent data forwarding and local resource coordination;

[0118] (2) Airborne UAV Coordination Module: Deployed on the UAV platform, it provides mobile perception, airborne forwarding and auxiliary computing capabilities when roadside infrastructure coverage is insufficient, event area is dynamically changing or local links are blocked, and collects information on the UAV's remaining energy, flight position, service load and link status.

[0119] (3) Cloud-based global control module: Deployed in the cloud computing center, used to perform cross-regional global policy maintenance, historical execution trajectory statistics, slow time scale protection parameter updates, and multi-regional collaborative scheduling management; (4) Workflow modeling and orchestration control module: Used to abstract the business to be processed into a directed acyclic graph. The workflow, combined with the computational requirements of task nodes, the data requirements of dependent edges, the node capability status, the link status, and the risk status, completes task mapping, resource allocation, protection configuration, and online recovery control.

[0120] The air-ground vehicle-to-everything (V2X) edge-cloud continuum collaborative execution environment proposed in this invention is as follows: Figure 1 As shown, it mainly consists of roadside edge servers, aerial drone nodes, a cloud computing center, and a workflow modeling and orchestration control unit. The roadside edge servers provide stable edge computing and communication access capabilities; the aerial drone nodes provide mobile sensing, aerial forwarding, and auxiliary computing capabilities; the cloud computing center is responsible for global policy maintenance and cross-regional scheduling management; and the workflow modeling and orchestration control unit abstracts the pending business into a directed acyclic graph. Workflow, and performs task mapping, resource allocation, protection configuration, and online recovery control.

[0121] The set of computing nodes in the air-to-ground vehicle network edge-cloud continuum is defined as follows:

[0122]

[0123] in, This represents the set of roadside edge servers. Represents a set of aerial drone nodes. This represents a cloud computing center; the communication connections between the nodes constitute a set of links. The pending business processes are modeled as a directed acyclic graph workflow. ,in, Represents a set of task nodes. Represents the set of task-dependent edges; for any task node Its computational requirements are defined as follows: For any dependent edge The amount of data to be transmitted is defined as Workflow release time is defined as The completion deadline is defined as .

[0124] Furthermore, in step 2), the network-level situational awareness is constructed as follows:

[0125] For any computing node Collect it in time slot The local observation vector is defined as:

[0126]

[0127] in, This represents the observation value of the communication quality of the link related to the node. This represents the observed value of the node's service processing capacity. This represents the observed value of packet transmission performance. Represents the observed value of the node's remaining energy state.

[0128] This represents the dangerous exposure observation value of the fault domain to which the node belongs.

[0129] To reduce the impact of transient disturbances on subsequent scheduling decisions, time smoothing is performed on the local observation vector to obtain the smoothed local situation parameters:

[0130]

[0131] in, This is the smoothing coefficient.

[0132] set up Represents a node In the time slot The set of communicating neighbors, then the node The fusion state quantity is represented as:

[0133]

[0134] in, Let be the neighborhood fusion weight, and satisfy:

[0135]

[0136] Based on the fused situational awareness of all nodes, construct the network-level situational awareness:

[0137]

[0138] in, This represents the calibrated estimate of the low-quantum link capability. This represents the calibrated estimate of low quantile service capacity. Represents the fault domain Risk rate estimate Indicates the total cost of the link. This represents an uncertainty indicator. Furthermore, the overall link cost is expressed as:

[0139]

[0140] in, , and These represent link latency cost, hazard cost, and maneuver cost, respectively.

[0141] This represents the corresponding weighting coefficient.

[0142] Figure 2 This invention illustrates a dual-timescale closed-loop orchestration framework. The framework comprises a situational awareness layer, a slow-timescale protection strategy update layer, a fast-timescale online orchestration and recovery layer, and an execution feedback layer. Specifically, the situational awareness layer is responsible for fusing link communication quality, node service capabilities, fault hazard exposure, and energy status to generate a network-level situational awareness. The slow-timescale protection strategy update layer adaptively updates checkpoint configuration, backup resource reservation, recovery mode preferences, and tail-end risk budgets based on the current network-level situational awareness and historical execution trajectories. The fast-timescale online orchestration and recovery layer performs primary execution node selection, backup node selection, path selection, communication resource allocation, computing resource allocation, and recovery action decisions for ready tasks based on the current time slot status. The execution feedback layer updates the queue status, risk status, energy status, and execution trajectory set based on the actual task execution results and feeds this information back to the protection strategy update process in subsequent time windows, thus forming a closed-loop mechanism where long-term protection configuration and short-term online recovery work together.

[0143] Furthermore, in step 3), the slow time-scale protection strategy is updated as follows:

[0144] In the window The protection strategy is defined as follows:

[0145]

[0146] in, Represents the set of protected dependent edges. Represents the periodic vector of checkpoints. Indicates tail risk budget, This represents the recovery mode preference vector. For any candidate dependency edge... and candidate checkpoint cycle Its protection cost is defined as:

[0147]

[0148] in, This indicates basic protection costs. Indicates the checkpoint cycle is Estimated protection overhead at that time

[0149] Display window The above is a scene dispersion statistic extracted from the execution trajectory. Correspondingly, the protection benefit is defined as:

[0150]

[0151] in, This indicates the amount of improvement in recovery latency. This indicates the amount of improvement in tail risk. This represents the improvement in survivability. Further, the slow timescale scoring function is defined as:

[0152]

[0153] Based on the scoring results of each candidate dependency edge under different candidate checkpoint periods, the checkpoint period with the highest score is selected, and the set of protected dependency edges is determined under the protection budget constraint. Furthermore, the window-level tail risk budget is updated as follows:

[0154]

[0155] in, Represents the interval projection operator. Display window Tail risk statistics within.

[0156] Furthermore, in step 4), the fast timescale online orchestration and online recovery method is as follows:

[0157] For any ready task Define the main execution node selection variable. And backup node selection variables ,satisfy:

[0158]

[0159]

[0160] in, Represents a node Selected as a mission The main execution node, Represents a node Selected as a mission Backup nodes. For dependent edges. If its transmission path is Then its transmission delay is:

[0161]

[0162] in, Indicates link The allocated transmission rate. If the task... At the node If executed above, its computation delay is:

[0163]

[0164] in, Indicates assignment to a node Service speed. Task The arrival time of the precursor is expressed as:

[0165]

[0166] Therefore, the task The completion time is expressed as:

[0167]

[0168] in, Indicates queuing delay. This indicates the recovery of additional latency. Further, the end-to-end completion latency of the workflow is expressed as:

[0169]

[0170] in, Indicates the completion time of the virtual exit node. Defines the recovery trigger indicator. for:

[0171]

[0172] in, Indicates the availability of the primary execution node. Indicates the remaining time limit margin for the task. Indicates the feasibility of contact; when At that time, an online recovery process is triggered, including backup switching, checkpoint replay, path rerouting, and task migration. For candidate actions... Its scoring function is defined as follows:

[0173]

[0174] in, This indicates the predicted execution and recovery costs. This indicates an increase in tail risk. The survivability gain is represented by minimizing the scoring function to determine the task orchestration result for the current time slot.

[0175] Furthermore, in step 5), the energy safety constraints, survivability constraints, and closed-loop feedback update methods are as follows:

[0176] The computing nodes in the system are divided into a set of fault domains. ,Task In the fault domain The deployment instruction quantity is defined as follows:

[0177]

[0178] Figure 3 This paper illustrates the fault domain modeling and task survivability assessment process in this invention. First, based on node geographic location, network topology, common-cause failure correlation, and environmental hazard exposure, the roadside edge server, aerial drone nodes, and related communication units in the system are divided into multiple fault domains. Second, the hazard rate of each fault domain is estimated, and the survival probability of each fault domain within the prediction time domain is calculated accordingly. Finally, based on the deployment of the task master replica in different fault domains, task-level survivability indicators are calculated, and it is further determined whether the corresponding workflow meets the preset survivability constraints.

[0179] Fault Domain In the prediction time domain The survival probability within is expressed as:

[0180]

[0181] Therefore, the task-level survivability metric is expressed as:

[0182]

[0183] The workflow is required to meet the following requirements:

[0184]

[0185] in, This represents the workflow survivability target threshold. For drone nodes... Its calculated energy consumption is expressed as:

[0186]

[0187] Its communication energy consumption is expressed as:

[0188]

[0189] The drone's remaining battery energy has been updated to:

[0190]

[0191] And satisfy:

[0192]

[0193] in, The energy safety threshold for drones.

[0194] 6) Joint optimization objectives and constraints:

[0195] The dual-time-scale orchestration method aims to minimize the long-term average completion delay of the workflow, and the objective function is expressed as:

[0196]

[0197] It also simultaneously satisfies constraints on computing capacity, bandwidth capacity, workflow survivability, deadline tail risk, and UAV energy safety.

[0198] 7) After each time slot or time window ends, the physical task queue, survivability virtual queue, tail risk virtual queue, UAV remaining energy status and execution trajectory set are updated according to the actual task execution results. The updated results are fed back to the protection strategy update process of the subsequent time window, thus forming a dual-timescale closed-loop orchestration mechanism that coordinates long-term protection configuration and short-term online recovery.

[0199] Simulation Examples

[0200] To verify the effectiveness of the proposed situational awareness-based dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks, performance was compared with several comparative methods in an air-to-ground V2X edge-cloud continuum simulation environment. The simulation environment included roadside edge servers, aerial drone nodes, and a cloud computing center. Service requests were represented by directed acyclic graphs with dependencies. The workflow is represented, and the performance of the method of the present invention is examined from aspects such as the number of workflow instances, workflow scale, communication-to-computation ratio, and number of edge nodes.

[0201] like Figure 4As shown, the performance of the method of the present invention and the comparative method are compared under different numbers of workflow instances. As the number of workflow instances increases, the degree of task contention and resource consumption in the system gradually increases, and the overall performance of each method changes to varying degrees. Compared with the comparative method, the method of the present invention can still maintain better task completion latency and more stable execution performance under high load conditions, indicating that the method of the present invention has better scheduling stability and load adaptability under varying numbers of workflow instances.

[0202] like Figure 5 As shown, the performance of the method of this invention and the comparative method are compared under different workflow scales. As the number of task nodes and the complexity of dependencies in a single workflow increase, the difficulty of task orchestration increases significantly. The method of this invention, by introducing network-level situational awareness and dual-timescale collaborative control, can effectively coordinate task mapping, resource allocation and online recovery processes when the workflow scale increases, thus demonstrating good scalability and complex task processing capabilities.

[0203] like Figure 6 As shown, the performance of the method of the present invention and the comparative method are compared under different communication-to-computation ratios. The change in the communication-to-computation ratio reflects the relative relationship between data transmission load and computation load during task execution; when the communication-to-computation ratio is high, the impact of link state fluctuations and bandwidth resource competition on task execution is more significant. The method of the present invention can dynamically adjust path selection, communication resource allocation, and protection strategies according to the network-level situation, thus maintaining good performance under different communication-to-computation ratios.

[0204] like Figure 7 As shown, the performance of the method of this invention is compared with that of the comparative method under different numbers of edge nodes. With the increase in the number of edge nodes, the available computing resources and parallel processing capabilities of the system are improved, but the coordination complexity between nodes also increases accordingly. The method of this invention can coordinately adjust the protection configuration and online orchestration strategy by combining the current network situation and historical execution trajectories, thus exhibiting better resource utilization and execution stability under different edge node scales.

Claims

1. A dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness, characterized in that... Includes the following steps: 1) Construct a collaborative execution environment for the air-ground vehicle-to-everything (V2X) network edge-cloud continuum. This collaborative execution environment includes roadside edge servers, aerial drone nodes, and a cloud computing center. The business to be processed is modeled as a directed acyclic graph with task dependencies. The workflow consists of a task node with corresponding computational requirements and a completion time limit, and a task dependency edge with corresponding data transmission requirements. 2) Collect local observation information of each computing node and communication link. The local observation information includes at least link communication quality, node service capability, fault hazard exposure degree and energy status. Generate network-level situational status through time smoothing, neighborhood fusion and uncertainty calibration, so as to serve as a unified decision basis for subsequent task orchestration and recovery control. 3) Based on the network-level situation and historical execution trajectory, update the protection strategy on a slow time scale. The protection strategy includes at least checkpoint configuration, backup resource reservation, recovery mode preference and tail risk budget of deadline, which are used to provide long-term protection parameters for task flow execution in subsequent time windows. 4) Based on the network-level situational awareness of the current time slot and the updated protection strategy, perform primary execution node selection, backup node selection, transmission path selection, communication resource allocation, and computing resource allocation for ready tasks on a fast time scale to form the online orchestration result of the current time slot; when it is detected that the primary execution node of the task is unavailable, the remaining time limit of the task cannot be met, or the communication contact relationship is invalid, trigger an online recovery process including backup switching, checkpoint replay, path rerouting, and task migration. 5) During mission execution and recovery, the computational energy consumption, communication energy consumption, and maneuvering energy consumption of the UAV nodes are considered simultaneously. When node capacity constraints, link capacity constraints, energy security constraints, or protection constraints are not met, resource projection and feasibility repair are performed on the orchestration results. Based on the actual mission execution results, the physical mission queue, survivability virtual queue, tail risk virtual queue, UAV remaining energy status, and execution trajectory set are updated, and the updated results are fed back to the protection strategy update process in the subsequent time window, thereby forming a dual-timescale closed-loop orchestration mechanism that coordinates long-term protection configuration and short-term online recovery.

2. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The air-ground vehicle-to-everything (V2X) edge-cloud continuum collaborative execution environment constructed in step 1) includes the following four parts: (1) Roadside edge service module, deployed on the roadside edge server, is used to provide low-latency computing, caching and communication access capabilities for vehicle terminals, roadside sensing devices and aerial drone nodes, and is responsible for undertaking some task node execution, task-dependent data forwarding and local resource coordination; (2) The aerial UAV collaboration module is deployed on the aerial UAV platform to provide mobile perception, aerial forwarding and auxiliary computing capabilities when the roadside infrastructure coverage is insufficient, the event area is dynamically changing or the local link is blocked, and to collect the UAV's remaining energy, flight position, service load and link status information. (3) Cloud-based global control module, deployed in the cloud computing center, is used to perform cross-regional global policy maintenance, historical execution trajectory statistics, slow time scale protection parameter updates, and multi-regional collaborative scheduling management; (4) Workflow modeling and orchestration control module, which is used to abstract the business to be processed into a directed acyclic graph (DAG) workflow, and combine the task node computing requirements, dependent edge data requirements, node capability status, link status and risk status to complete task mapping, resource allocation, protection configuration and online recovery control.

3. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The network-level situational awareness described in step 2) is constructed as follows: 1) Define the computing nodes in the air-ground vehicle-to-everything (V2X) edge-cloud continuum as a node set. Its expression is: in, This represents the set of roadside edge servers. Represents a set of aerial drone nodes. This represents a cloud computing center; the communication connections between the nodes constitute a set of links. The pending business processes are modeled as a directed acyclic graph workflow. ,in, Represents a set of task nodes. Represents the set of task-dependent edges; for any task node Its computational requirements are defined as follows: For any dependent edge The amount of data to be transmitted is defined as . 2) For any computing node Collect it in time slots The local observation vector is defined as: in, This represents the observation value of the communication quality of the link related to the node. This represents the observed value of the node's service processing capacity. This represents the observed value of packet transmission performance. Represents the observed value of the node's remaining energy state. This represents the dangerous exposure observation value of the fault domain to which the node belongs. 3) Perform time smoothing on the local observation vector to obtain the smoothed local situation variables: in, This is the smoothing coefficient. 4) Let Represents a node In the time slot The set of communicating neighbors of a node The fusion state quantity is represented as: in, Let be the neighborhood fusion weight, and satisfy: 5) Construct the network-level situational awareness based on the fused situational awareness of all nodes: in, This represents the calibrated estimate of the low-quantum link capability. This represents the calibrated estimate of low quantile service capacity. Represents the fault domain Risk rate estimate Indicates the total cost of the link. Indicators representing uncertainty.

4. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The slow timescale protection strategy update method described in step 3) is as follows: 1) Within the time window The protection strategy is defined as follows: in, Represents the set of protected dependent edges. Represents the periodic vector of checkpoints. Indicates tail risk budget, This represents the recovery mode preference vector. 2) Based on the network-level situation status and historical execution trajectory within the current time window, evaluate the protection cost and protection benefit of candidate dependency edges under different candidate checkpoint cycles. The protection cost is at least related to the resource consumption, protection overhead and scenario uncertainty caused by checkpoint setting, and the protection benefit is at least related to the improvement in recovery latency, the improvement in tail risk of deadline, and the improvement in task survivability. 3) Determine the protection priority corresponding to each candidate dependency edge based on the protection cost and the protection benefit, and select the checkpoint period, determine the set of protected dependency edges and the corresponding backup resource reservation method according to the protection priority, so as to form a slow time scale protection strategy within the current time window; 4) Update the tail risk budget based on the window-level tail risk statistics. The updated tail risk budget is as follows: in, Represents the interval projection operator. Display window Tail risk statistics within. 5) The updated set of protected dependent edges, checkpoint period vector, tail risk budget, and recovery mode preference vector are used as constraints and decision inputs for subsequent fast timescale online orchestration and online recovery, so as to enable long-term protection parameters to guide and constrain short-term task scheduling behavior.

5. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The fast timescale online orchestration and online recovery method described in step 4) is as follows: 1) For any ready task Define the main execution node selection variable. And backup node selection variables ,satisfy: in, Represents a node Selected as a mission The main execution node, Represents a node Selected as a mission Backup nodes. 2) For dependent edges If its transmission path is Then its transmission delay is: in, Indicates link The allocated transmission rate. If the task... At the node If executed above, its computation delay is: in, Indicates assignment to a node Service speed. Task The arrival time of the precursor is expressed as: Therefore, the task The completion time is expressed as: in, Indicates queuing delay. This indicates the recovery of additional latency. Further, the end-to-end completion latency of the workflow is expressed as: in, This indicates the completion time of the virtual exit node. 3) Construct a recovery trigger indicator based on the availability of the task's main execution node, the remaining time margin of the task, and the feasibility of contact: in, Indicates the availability of the primary execution node. Indicates the remaining time limit margin for the task. Indicates the feasibility of contact; when When this occurs, an online recovery process is triggered, including backup switching, checkpoint playback, path rerouting, and task migration. 4) The candidate actions include at least the selection of the primary execution node, the selection of the backup node, the selection of the transmission path, the allocation of communication resources, the allocation of computing resources, and the selection of recovery actions. The scoring results are constructed based on the predicted execution and recovery costs, the tail risk increment, and the survivability gain to select the best task scheduling result for the current time slot. 5) After obtaining the task orchestration result for the current time slot, perform a feasibility check. If there are resource conflicts, protection conditions are not met, or recovery actions are not executable, correct the task orchestration result to ensure that the output online orchestration result meets the task execution requirements of the current time slot.

6. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The fault domain modeling and task survivability constraint methods described in step 5) are as follows: 1) Divide the computing nodes in the air-to-ground vehicle-to-everything (V2X) edge-cloud continuum into a set of fault domains. Computational nodes within the same fault domain share common causes of fault correlation, and different fault domains correspond to different levels of hazard exposure and failure propagation characteristics. 2) For any task Define it in the fault domain The deployment instructions on the above are: in, Indicates task At the node The main execution node selects variables. Indicates task At the node On the backup node selection variable; when When, it indicates a task. At least one of the primary replicas is deployed in the fault domain. ; 3) When the current task deployment scheme does not meet the workflow survivability constraints, adjust the main execution node, backup node or recovery action to ensure that the updated task orchestration results meet the workflow-level survivability requirements. 4) Based on the task-level survivability index of each task in the workflow, construct workflow survivability constraints, expressed as follows: in, Represents workflow The set of task nodes This represents the target threshold for workflow survivability. 5) When the current task deployment scheme does not meet the workflow survivability constraints, adjust the main execution node, backup node or recovery action to ensure that the updated task orchestration results meet the workflow-level survivability requirements.

7. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The energy security constraint method for the aerial drone node mentioned in step 5) is as follows: 1) For any aerial drone node Its calculated energy consumption is expressed as: in, Indicates drone node The equivalent switching energy consumption coefficient, Indicates task The computational requirements This indicates the service rate allocated to the drone node. This represents the static power of the drone node; 2) For any dependent edge In the time slot The corresponding communication energy consumption is expressed as follows: in, Indicates dependent edges The amount of data to be transmitted Indicates link The allocated transmission rate; 3) For any aerial drone node In the time slot The energy consumption related to the tasks undertaken internally is expressed as follows: in, Indicates time slot The internal workflow collection, Represents workflow In the time slot The set of ready tasks, This indicates the reduction factor for backup reserved resources; 4) The remaining battery energy of the aerial drone node u is updated as follows: in, Indicates drone node In the time slot The energy consumption of the vehicle; 5) The aerial unmanned aerial vehicle (UAV) node satisfies the following energy security constraints: in, This is the energy safety threshold for the drone. When the remaining battery energy of an aerial drone node falls below this energy safety threshold, the aerial drone node is restricted from undertaking new primary or backup tasks.

8. The dual-timescale orchestration method for air-to-ground vehicle-to-everything (V2X) networks based on situational awareness as described in claim 1, characterized in that... The dual-time-scale orchestration method aims to minimize the long-term average completion delay of the workflow, and combines capacity constraints, bandwidth constraints, tail risk constraints, and closed-loop feedback updates for joint optimization. Specifically, it includes the following: 1) During task orchestration, the following constraints are met: computing capacity constraint, bandwidth capacity constraint, workflow survivability constraint, deadline tail risk constraint, and airborne UAV node energy safety constraint. The computing capacity constraint is used to limit the task load that each computing node can undertake in the current time slot. The bandwidth capacity constraint is used to limit the allocable transmission resources of each link and node. The deadline tail risk constraint is used to limit the risk level of the workflow completion delay exceeding its deadline. 2) After each time slot or time window ends, update the physical task queue, survivability virtual queue, tail risk virtual queue, remaining energy status of aerial UAV nodes, and execution trajectory set according to the actual task execution results. The execution trajectory set includes at least the task execution delay, recovery trigger status, resource usage status, and risk statistics results. 3) Feed the update results back to the protection strategy update process in the subsequent time window to achieve a closed-loop feedback update between slow time scale protection strategy update and fast time scale online orchestration and online recovery; 4) The closed-loop feedback update is used at least to correct the protection parameters, resource allocation strategies, and recovery action selection results within subsequent time windows, thereby improving the complexity... The execution stability, recoverability, and time-limit fulfillment capability of workflows in dynamic air-ground collaborative environments.

Images