Air-ground integration-oriented deterministic edge computing power network unmanned aerial vehicle auxiliary deployment and task scheduling method

Abstract

The invention discloses an air-ground integration-oriented deterministic edge computing power network unmanned aerial vehicle auxiliary deployment and task scheduling method. The method comprises the steps of constructing a multi-layer computing power network model based on network topology; constructing a DAG graph model according to the calculation task; generating an optimal network forwarding path for each workflow of the DAG model in the multi-layer computing power network model, and completing path planning, time slot scheduling and subtask unloading of the workflows; a vacant deployment position of a computing power calculation node is positioned through a multi-layer computing power network model, matching of the unmanned aerial vehicle and the vacant deployment position is completed in combination with a subtask unloading result and a computing power demand of the subtask, and a service time slot and a service deployment position of the unmanned aerial vehicle are adjusted; all task transmission is completed in a determined service time slot period based on the service time slot and the service deployment position of the unmanned aerial vehicle and through time-aware gating control. The task is ensured to be completed in a controllable time delay range, and the overall delivery efficiency of the task is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of communication networks, and more specifically, to a method for deploying and task scheduling of an unmanned aerial vehicle (UAV) assisted for a deterministic edge computing power network oriented to air-ground integration. Background Art

[0002] With the continuous development and maturity of new network services and applications, cloud computing, edge computing, and intelligent terminal devices have developed rapidly, and computing resources show a trend of ubiquitous deployment. In the face of challenges such as large network scale, scattered resources, and heterogeneity, in this context, the concept of a computing power network (CPN) has been proposed. The computing power network deeply integrates heterogeneous network resources, computing resources, storage resources, etc., coordinates distributed computing resources, realizes extensive computing power interconnection, and provides computing power services that meet user needs. The air-ground integrated network is a new communication system formed by combining a lifting platform (tethered balloon or UAV) carrying communication modules or devices with a ground base station. Due to its advantages such as flexible deployment and ultra-long coverage, the air-ground integrated network can form a heterogeneous network structure with overlapping air-ground coverage. UAV-assisted edge computing and communication have many natural advantages such as flexible deployment and strong mobility. By utilizing the flexible movement characteristics of UAVs and equipping them with computing servers with certain computing and processing capabilities, high-energy efficiency and low-latency auxiliary processing of terminal device computing tasks can be realized. Time-sensitive Networking (TSN) is a new type of deterministic network with characteristics such as low latency and low jitter, and can meet the requirements of modern network transmission control. Based on time synchronization, TSN solves problems such as the determinism and reliability of Ethernet data transmission through mechanisms such as shaping scheduling and reliable redundancy, and has become a highly competitive new generation of deterministic network solutions with advantages such as open standards, deterministic latency, and service guarantee.

[0003] With the expansion of intelligent applications and scenarios, the tasks processed within a computing power node are not of a single type or fixed size. Facing differentiated tasks with high computational intensity and real-time requirements, it is impossible to achieve multi-resource collaborative optimization based on existing architectures and scheduling technologies to realize computing power scheduling and deployment according to service requirements. In addition, for computationally intensive tasks, there are problems such as insufficient single-point computing power and uncontrollable task delivery time. Although the processing and computing time of tasks have been evaluated, no mechanism and method for resource reservation have been configured, which may lead to long task computing time. Even if a processor with sufficient computing resources is called, the computing power node may be occupied for a long time, resulting in too long a queuing time for the computing queue and affecting the timely delivery of other tasks. In addition, the uncertainty of network transmission will also cause large fluctuations in the end-to-end delivery time of workflows, thus affecting the task delivery time.

[0004] Therefore, how to integrate the computing power network, time-sensitive network, and air-ground integrated architecture to provide a task scheduling method with drone-assisted deployment, transmission, and processing determinacy to solve the above problems. Summary of the Invention

[0005] In view of this, the present invention provides a method for drone-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration. The present invention makes an integrated decision on computing power resource scheduling, generation of gating lists, and task injection network control, ensuring deterministic transmission of the workflow while efficiently utilizing multi-point computing power resources and improving the overall task delivery efficiency.

[0006] At the same time, the present invention proposes a multi-layer computing power network model based on drone-assisted deployment. The drone-assisted deployment improves the flexibility of network deployment, supports decomposing a complex task into multiple subtasks, offloading them to multiple computing power nodes for processing, and then achieving transmission within a determined time slot period through precise time-aware gating control, thereby ensuring the goal of controllable task delivery time as a whole.

[0007] To achieve the above objectives, the present invention adopts the following technical solutions:

[0008] A method for drone-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration, including:

[0009] Constructing a multi-layer computing power network model based on the network topology structure;

[0010] Constructing a DAG graph model according to the computing task;

[0011] Generating an optimal network forwarding path for each workflow of the DAG graph model in the multi-layer computing power network model to complete the path planning, time slot scheduling, and subtask offloading results of the workflow;

[0012] Locating the vacant deployment positions of the computing power nodes through the multi-layer computing power network model, matching the drones with the vacant deployment positions in combination with the subtask offloading results and the computing power requirements of the subtasks, and adjusting the service time slots and service deployment positions of the drones;

[0013] Completing drone-assisted deployment based on the service time slots and service deployment positions of the drones, and achieving all task transmissions within a determined service time slot period through time-aware gating control.

[0014] Preferably, the network topology structure includes computing power nodes, sensing nodes, task initiation ends, and task result receiving ends. The sensing nodes, task initiation ends, and task result receiving ends are connected to the computing power nodes through Ethernet links, and the computing power nodes include computing power calculation nodes and computing power routing nodes.

[0015] Preferably, a multi-layer computing power network model is constructed based on the network topology structure, specifically including:

[0016] Model the attributes of computing power nodes and sensing nodes according to physical resources;

[0017] Assume that each computing power routing node is connected to a computing power calculation node, and establish edges according to the node layering and the physical connections between the actual nodes that actually exist and are deployed within the scenario. The edges represent communication links for transmitting data, and the actual nodes include computing power nodes and sensing nodes.

[0018] Preferably, a DAG graph model is constructed according to the computing tasks, specifically including:

[0019] The task initiation end creates a computing task according to the user requirements or the instructions of the upper-layer application and sends it to the computing network brain;

[0020] The computing network brain completes the modeling of the computing task, decomposes the computing task into multiple subtasks, and there are data dependency relationships and sequential dependency relationships between the subtasks;

[0021] Abstract each subtask into a node of a directed acyclic graph to determine the task nodes;

[0022] Determine the directed edges of the directed acyclic graph according to the task sequence and data dependency relationships;

[0023] Determine the DAG graph model based on the task nodes and directed edges.

[0024] Preferably, the constraint conditions for path planning and time slot scheduling include:

[0025] a) Each subtask is offloaded to different computing power calculation nodes;

[0026] b) According to the task sequence, the processing result of the previous subtask is the input stream of the next subtask;

[0027] c) Routing constraint: For the intermediate nodes through which the traffic flows, the incoming port traffic is equal to the outgoing port traffic;

[0028] d) The total task completion delay does not exceed its tolerable maximum delay;

[0029] e) The workflow is transmitted following the path sequence;

[0030] f) The computing resources reserved for the subtask are sufficient for it to complete processing within one time slot;

[0031] g) The time when the workflow generated by the first subtask is injected into the network does not exceed its task cycle;

[0032] h) The number of workflows planned within a unified time slot does not exceed the queue length;

[0033] i) CQF-3 mechanism constraint: Check whether the data packet arrives within the right end of the window, ensure that the workflows in the odd queue are sent from the upstream node within the odd cycle loop and are received by the downstream node and cached in the even queue within the same time slot cycle. If the data packet does not arrive within the corresponding cycle, place it in the buffer queue and forward it to the odd queue in the next time slot cycle.

[0034] Use the Z3 solver to solve and obtain the final optimization results, getting the offloading location of subtasks, the path planning of workflows, and the time slot scheduling results of workflows.

[0035] Preferably, through time-aware gating control, all task transmissions are completed within the determined service time slot cycle, specifically including:

[0036] After placing the workflow into the receiving queue, determine whether to process it in the local computing queue or directly forward it through the data transmission queue.

[0037] If it is processed in the local computing queue, place it into the corresponding local computing queue according to the priority of the workflow, allocate predetermined computing resources and storage resources for the task according to resource requirements, control the processing order and time of local computing tasks, and send and store them in the data transmission queue within the specified service time slot.

[0038] If it is processed through the data transmission queue, judge and enter the corresponding data transmission queue according to the priority of the workflow, and forward the workflow to the next node within the specified service time slot.

[0039] Preferably, both the local computing queue and the data transmission queue support 8 priorities.

[0040] Preferably, the forwarding principles of the data transmission queue and the local computing queue according to priority are:

[0041] Q in the gating list GCL 5、 Both Q6 and Q7 adopt the CQF-3 queue model. The CQF-3 queue model periodically opens and closes the queue gating in units of T. Q7 is closed in odd time slots and opened in even time slots. Q6 is opened in odd time slots and closed in even time slots. Q5 only receives data packets that do not arrive at Q7 and Q6 within the specified time slots. The gating of Q5 is connected to the gating of Q7 and Q6 in the open state. If the gating corresponding to the original outgoing queue is open, forward it; if it is closed, continue to cache it in Q5.

[0042] The queues Q0 to Q4 in the gating list GCL adopt the priority queue model and are scheduled in sequence according to the priority order. Workflows with priorities from 0 to 4 are respectively placed into the queues Q0 to Q4, and only allowed to be sent when the gating is open.

[0043] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration. This method uses UAVs with strong mobility and flexible deployment as an expansion of computing resources in the deterministic computing power network. By designing effective UAV deployment methods and task scheduling methods, the overall system performance can be improved to meet diverse task requirements. Based on the computing power network technology, various resources within the system (including computing, network, storage, AI models, etc.) are uniformly measured and scheduled to achieve optimal resource allocation. In addition, through the enhancement of time-sensitive technology, based on the CQF-3 mechanism, by planning the time slots for workflow injection into the network, it can ensure that tasks are completed within a controllable time delay range. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0045] Figure 1 It is a flowchart of the method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration provided by the present invention.

[0046] Figure 2 It is a schematic diagram of the scenario of a deterministic edge computing power network for air-ground integration provided by the present invention.

[0047] Figure 3 It is an architecture diagram of the multi-layer computing power network model provided by the present invention.

[0048] Figure 4 It is a structure diagram of the DAG graph model provided by the present invention.

[0049] Figure 5 It is a schematic diagram of task decomposition and scheduling provided by the present invention.

[0050] Figure 6 It is a structure diagram of the internal queue of the computing power node provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0052] An embodiment of the present invention discloses a method for drone-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration, as Figure 1 and Figure 2 shown, including:

[0053] Construct a multi-layer computing power network model based on the network topology structure;

[0054] Construct a DAG graph model according to the computing tasks;

[0055] Generate an optimal network forwarding path for each workflow of the DAG graph model in the multi-layer computing power network model to complete the path planning, time slot scheduling, and subtask offloading results of the workflow;

[0056] Locate the vacant deployment positions of the computing power nodes through the multi-layer computing power network model. Combine the subtask offloading results and the computing power requirements of the subtasks, and use the greedy matching algorithm to complete the matching of the drones and the vacant deployment positions. Considering the resource requirements of the tasks and the available computing power resources of the drones, for the cumulative task resource requirements and the satisfaction of the drone resource computing resources within a single time slot, each time select the drone resource that can meet the task requirements to the greatest extent, and adjust the service time slot and service deployment position of the drone; Here, the vacancy is because the number and positions of the computing power nodes in the multi-layer computing power network model are constructed based on the computing power routing nodes and the computing power nodes in the actual scenario. In the constructed multi-layer computing power network model, the computing power nodes may exist or may not exist. For the positions where they do not exist, dynamic tethered drones need to be deployed as computing power nodes to complete the calculation of the tasks;

[0057] Complete the drone-assisted deployment based on the service time slot and service deployment position of the drone, and through time-aware gating control, realize the completion of all task transmissions within a determined service time slot period.

[0058] In this embodiment, the network topology includes computing power nodes, sensing nodes, task initiators, and task result receivers. The sensing nodes, task initiators, and task result receivers are connected to the computing power nodes through Ethernet links. The task initiator is the starting point of the computing power resource scheduling and task processing flow. It is responsible for initiating various computing tasks, creating specific computing tasks according to user requirements or instructions from upper-layer applications, and providing detailed feature descriptions of the generated tasks, including information such as the type of task (e.g., compute-intensive, data-intensive), task priority, and the task's data dependency. The task initiator first sends the computing task information to the computing and networking brain. The computing and networking brain performs resource allocation and computing power scheduling for the task based on the global resource distribution and task information. The data required for the computing task is generated by terminal devices (within the sensor domain) and is analyzed and processed through the collaboration of multiple computing power nodes. Finally, the result of the task is sent to the task receiver. The task receiver is responsible for receiving the task result, and this role can be the same user as the task initiator. The computing power nodes are further divided into computing power calculation nodes and computing power routing nodes according to their own functions. The drone computing power nodes serve as extended computing resources for the computing power calculation nodes and are used to assist in task processing, computing, and analysis. The computing power calculation node is the unit that actually executes computing tasks and is responsible for receiving tasks and performing data processing, model operations, etc. The computing power routing node connects each computing power calculation node within the deterministic computing power network to ensure that the workflow and data can be accurately and efficiently transmitted to the appropriate computing power calculation node for execution.

[0059] Construct a multi-layer computing power network model, as Figure 3 shown, specifically including:

[0060] Model the attributes of computing power nodes and sensing nodes based on physical resources. The physical resources include computing power node physical resources and sensing node physical resources. The computing power node physical resources include computing resources (including CPU, GPU, TPU), storage resources, network resources, and algorithm resources, etc. The sensing node physical resources include data acquisition modules, communication resources, energy and endurance, local computing resources, etc.;

[0061] Assume that each computing power routing node is connected to a computing power calculation node. Establish edges based on node layering and the physical connections between the actual nodes that exist and are deployed within the scenario. The edges represent communication links for transmitting data. The actual nodes include computing power nodes and sensing nodes. Among them, v1 - v6 represent computing power calculation nodes, w1 - w6 represent computing power routing nodes, and s1 - s8 represent sensing nodes.

[0062] In this embodiment, construct a DAG graph model according to the computing task, as Figure 4 shown, specifically including:

[0063] The task initiation end creates a computing task according to the user's requirements or the instructions of the upper-layer application, and sends the computing task to the computing network brain. The computing network brain completes the modeling of the computing task, decomposes the computing task into multiple subtasks, and there are data dependence relationships and sequential dependence relationships between the subtasks;

[0064] Abstract each subtask as a node of a directed acyclic graph (DAG), determine the task nodes. The circles in the graph represent a node of the DAG. For example, tasks such as vehicle recognition, license plate representation extraction, and trajectory prediction in vehicle target recognition are all regarded as separate nodes. Each task node carries task-related attributes, including task ID, subtask type, subtask name, priority, estimated execution time, resource requirements, etc.;

[0065] Determine the directed edges of the directed acyclic graph according to the task sequential dependence relationship; the directed edges represent the dependence relationship between tasks, pointing from one dependent task to another dependent task, that is, the start of a certain subtask must depend on the completion of another subtask, and the arrow direction represents the execution order. This structure ensures the sequentiality and rationality of task execution, and avoids logical errors caused by circular dependencies. The flow on the directed edge is generated by the subtask and will be used as the input data of the next subtask. The present invention is called the workflow transmitted at the computing power routing node, that is, f1 to f in the graph 16 . The workflow carries workflow-related attributes, including priority, traffic type, data transmission volume, period, etc.

[0066] Determine the DAG graph model based on the task nodes and directed edges, ensure that there are no loops in the graph, and there is no situation of mutual dependence between tasks. Nodes without direct dependence relationships can be executed in parallel.

[0067] After completing the modeling of the DAG graph model, the sensing node forwards the data related to the computing task to the computing power calculation node through the computing power routing node. The computing power calculation node processes the computing task, generates partial results, encapsulates them into the workflow, and forwards them to the next computing power calculation node through the computing power routing node until all tasks are processed.

[0068] The present invention assumes that each computing power node in the network can only process one subtask at a time, and the task is not interrupted during execution. The computing network brain comprehensively controls the information of various resources within the system, including computing power resources (such as CPUs, GPUs, NPUs, etc.), network resources (such as link bandwidth, network congestion, etc.), and dynamically allocates resources according to task requirements to ensure the efficient utilization of resources. At the same time, it monitors the performance status of each node and link in real time to ensure that faults and congestion locations can be quickly located and repair measures can be taken. The multi-layer computing power network model connects computing power nodes to each computing power routing node. Assuming a one-to-one connection relationship between computing power routing nodes and computing power nodes, on the physical topology of the deterministic edge computing power network, the deployment of tethered drones ensures that the number of computing power routing nodes and computing power nodes is the same and has a one-to-one association relationship. The scheduling of any complex task deconstructed based on the DAG model can be represented by a path sequence within this model.

[0069] Associate the sensing nodes according to the data source required by the task, so as to Figure 5 Taking vehicle target detection as an example, the computing power routing nodes connected to the sensors associated with this task are different from the computing power routing nodes associated with the task receiving end. This task contains 4 subtasks, labeled t1, t2, t3, and t4. The sensing node associated with this task is s1, which is used to collect vehicle information in this area. Based on the DAG graph model, the following graph is obtained, and the association relationship between subtasks is analyzed to judge the sequence, dependency, or parallel relationship of subtasks. Vehicle target detection involves four subtasks: vehicle target detection data collection (vehicle) t1, vehicle recognition t2, license plate identification extraction and trajectory prediction t3, and detection result (transmission back) t4. Among them, t1 and t4 do not require task calculation and need to complete data transmission tasks, while t2 and t3 need to complete task offloading and calculation on the computing power nodes.

[0070] The execution order of the subtasks is t1, t2, t3, t4, and there are data and sequence dependencies between the tasks. f1 is the workflow from t1 to t2, which needs to transmit the collected vehicle information to the computing power node processing t2 to complete the task calculation. Similarly, after vehicle recognition, the vehicle recognition data is transmitted to the computing power node processing t3 through the workflow f2, and so on, to complete task parsing, deconstruction, and modeling. After resource evaluation and computing power scheduling, the positions of the computing power nodes where t2 and t3 are offloaded are v1 and v5, and the path sequence is [w1, v1, w3, w5, v5, w6].

[0071] In this embodiment, an optimal network forwarding path is generated for each workflow of the DAG graph model in the multi-layer computing power network model to complete the path planning, time slot scheduling, and sub-task offloading results of the workflow. The joint allocation problem of task offloading and workflow transmission is established as a routing planning and time slot scheduling problem, and the path with the fewest total hops of consecutive workflows with dependencies is preferentially selected.

[0072] The decision variables of the routing planning and time slot scheduling problem are the traffic routing scheduling variable and the traffic time slot scheduling variable.

[0073] The constraint conditions of the routing planning and time slot scheduling problem include:

[0074] a) Each sub-task is offloaded to different computing power nodes;

[0075] b) According to the sequence of tasks, the processing result of the previous sub-task is the input stream of the next sub-task;

[0076] c) Routing constraint: For the intermediate nodes through which the traffic flows, the in-port traffic is equal to the out-port traffic;

[0077] d) The total task completion delay does not exceed its tolerable maximum delay;

[0078] e) The workflow is transmitted in accordance with the path sequence;

[0079] f) The computing resources reserved for the sub-task are sufficient for it to complete processing within one time slot;

[0080] g) The time when the workflow generated by the first sub-task is injected into the network does not exceed its task cycle;

[0081] h) The number of workflows planned within the unified time slot cannot exceed the queue length;

[0082] i) CQF-3 mechanism constraint: Check whether the data packet arrives within the right end of the window, ensure that the workflows in the odd queue are sent from the upstream node within the odd cycle loop, and are received by the downstream node and cached in the even queue within the same time slot cycle. If the data packet does not arrive within the corresponding cycle, it is placed in the buffer queue and forwarded to the odd queue in the next time slot cycle;

[0083] The Z3 solver is used to solve and obtain the final optimization result, and the offloading location of the sub-task, the routing result of the workflow, and the time slot scheduling result of the workflow are obtained.

[0084] In this embodiment, Figure 6 For the internal queue structure diagram of the computing power node, the computing power node receives the workflow ( Figure 4 f1 to f within 16After being placed into the receiving queue, the decision-making module needs to determine whether to process it in the local computing queue or directly forward the data through the data transfer queue. Both the data transfer queue and the local computing queue support 8 priority queues, and a gating module is configured in the data transfer queue. If it is a direct data forwarding, the decision-making module judges according to the priority of the workflow, places the workflow into the corresponding data transfer queue, and forwards the workflow within the specified time slot. If it is processed in the local computing unit, the decision-making module places it into the corresponding local computing queue according to the priority of the workflow, allocates predetermined computing resources and storage resources for the task according to the resource requirements of the task, precisely controls the processing sequence and time of the local computing task, and then sends and stores it into the data transfer queue within the specified time slot. According to the configured gating list (GCL), an opening window is provided in a specific time slot to forward the workflow. The transmission selector sequentially selects the workflows to reach the output port according to the priority, checks the integrity of the data frame at the output port, and finally sends it to the next node through the physical layer (depending on its routing, it may be forwarded to the computing power calculation node or the computing power routing node). In the data transfer queue, Q 5、 Q6 and Q7 adopt the CQF-3 queue model. Based on the CQF (cyclic queuing and forwarding) mechanism, a third buffer queue is added to avoid packet loss when there is an error in the arrival time slot of the data frame, temporarily store the packets that are not forwarded in time, and forward them to the next node when the gating corresponding to the original output queue is opened. The CQF-3 queue periodically opens and closes the queue gating in units of T. Q7 is closed in odd time slots and opened in even time slots, and Q6 is opened in odd time slots and closed in even time slots. Q5 only receives the packets that do not reach Q7 and Q6 in the specified time slot. The gating of Q5 is connected to the gating of Q7 and Q6 in the open state. If the gating corresponding to the original output queue is opened, it will be forwarded; if it is closed, it will continue to be buffered in Q5. That is, if the data frame was originally arranged in Q7 or Q6, since the data frame did not arrive according to the time slot regulations, it is temporarily stored in Q5 and forwarded from the queue when the gating of Q7 or Q6 queue is opened. The queues Q0 to Q4 adopt the strict priority queue model and are scheduled in the order of priority (0 < 1 < 2 < 3). The workflows with priorities 0 to 4 are respectively placed into the queues Q0 to Q4, and are only allowed to be sent when the gating is opened. Corresponding to the data transfer queue, the local computing queue adopts the same queue model, and places the workflow generated after the processor completes the task processing into the data transfer queue with the same priority, waiting to be forwarded.

[0085] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.

[0086] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration, characterized in that Including: Construct a multi-layer computing power network model based on the network topology structure; Construct a DAG graph model according to the computing task; Generate the optimal network forwarding path for each workflow in the DAG graph model in the multi-layer computing power network model, and complete the path planning, time slot scheduling and subtask offloading results of the workflow; Locate the vacant deployment positions of the computing power nodes through the multi-layer computing power network model, combine the subtask offloading results and the computing power requirements of the subtasks, complete the matching of the drones with the vacant deployment positions, and adjust the service time slots and service deployment positions of the drones; Complete the drone-assisted deployment based on the service time slots and service deployment positions of the drones, and through time-aware gate control, realize the completion of all task transmissions within the determined service time slot period.

2. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 1, wherein The network topology structure includes computing power nodes, sensing nodes, task initiators and task result receivers. The sensing nodes, task initiators and task result receivers are connected to the computing power nodes through Ethernet links. The computing power nodes include computing power calculation nodes and computing power routing nodes.

3. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 2, wherein Construct a multi-layer computing power network model based on the network topology structure, specifically including: Model the attributes of the computing power nodes and sensing nodes according to the physical resources; Assume that each computing power routing node is connected to a computing power calculation node, and establish edges according to node layering and the physical connections between the actual nodes that actually exist and are deployed in the scenario. The edges represent the communication links for transmitting data. The actual nodes include computing power nodes and sensing nodes.

4. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 2, wherein Construct a DAG graph model according to the computing task, specifically including: The task initiator creates a computing task according to the requirements and sends it to the computing network brain; The computing network brain completes the modeling of the computing task, disassembles the computing task into multiple subtasks, and there are data dependence relationships and sequential dependence relationships between the subtasks; Abstract each subtask as a node of a directed acyclic graph to determine the task nodes; Determine the directed edges of the directed acyclic graph according to the task sequence and data dependence relationship; Determine the DAG graph model based on the task nodes and directed edges.

5. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 4, wherein, The constraint conditions for the path planning and time slot scheduling problems include: a) Each subtask is offloaded to a different computing power calculation node; b) According to the task sequence, the processing result of the previous subtask is the input stream of the next subtask; c) For the intermediate nodes through which the traffic flows, the incoming port traffic is equal to the outgoing port traffic; d) The total task completion delay does not exceed its tolerable maximum delay; e) The workflow is transmitted in accordance with the path sequence; f) The computing resources reserved for the subtasks are sufficient for them to complete processing within one time slot; g) The time when the workflow generated by the first subtask is injected into the network does not exceed its task cycle; h) The number of workflows planned within the same time slot cannot exceed the queue length; i) CQF-3 mechanism constraint: Check whether the data packet arrives within the right end of the window, ensure that the workflows in the odd queue are sent from the upstream node within the odd cycle loop, and are received by the downstream node and cached in the even queue within the same time slot cycle. If the data packet does not arrive within the corresponding cycle, place it in the buffer queue and forward it to the odd queue in the next time slot cycle; The final optimization result is obtained by using the Z3 solver, and the offloading location of subtasks, the path planning of the workflow, and the time-slot scheduling result of the workflow are obtained.

6. The method for drone-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 1, wherein Through time-aware gating control, all task transmissions are completed within a determined service time-slot period, specifically including: After the workflow is placed in the receive queue, the decision module determines whether to process it in the local computing queue or directly forward it through the data transmission queue; If it is processed in the local computing queue, it is placed in the corresponding local computing queue according to the priority of the workflow, and the predetermined computing resources and storage resources are allocated to the tasks according to the resource requirements. The processing order and time of local computing tasks are controlled, and it is sent and stored in the data transmission queue within the specified service time-slot; If it is processed through the data transmission queue, it is judged to enter the corresponding data transmission queue according to the priority of the workflow, and the workflow is forwarded to the next node within the specified service time-slot.

7. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 6, wherein Both the local computing queue and the data transmission queue support 8 priorities.

8. The method for UAV-assisted deployment and task scheduling of a deterministic edge computing power network for air-ground integration according to claim 7, wherein The forwarding principle of the data transmission queue and the local computing queue according to priority is: Q in the gating list GCL 5、 Both Q6 and Q7 adopt the CQF-3 queue model. The CQF-3 queue model periodically alternates the opening and closing of the queue gate in units of T. Q7 is closed in odd time slots and opened in even time slots. Q6 is opened in odd time slots and closed in even time slots. Q5 only receives the data packets that do not arrive at Q7 and Q6 at the specified time slots. The gating of Q5 is connected to the gating of Q7 and Q6 in the open state. If the gating corresponding to the original outgoing queue is open, it is forwarded; if it is closed, it continues to be cached in Q5; For the queues Q0 to Q4 in the gating list GCL, the priority queue model is adopted, and they are scheduled in sequence according to the priority order. The workflows with priorities 0 to 4 are placed in the queues Q0 to Q4 respectively, and only allowed to be sent when the gating is open.

Images