Edge-edge cooperation system and method supporting task awareness and resource sharing

Abstract

The invention discloses an edge-edge cooperation system and method supporting task awareness and resource sharing, and relates to the technical field of edge computing, and the method comprises the following steps: a plurality of edge nodes periodically collect resource state information of the nodes, and generate state vectors of a uniform format; meanwhile, a newly arrived task is described as structured task information; calculating a comprehensive score value of each edge node based on the state vector; dynamically selecting an optimal target node from the plurality of edge nodes based on a multi-factor scoring mechanism and an anti-jitter strategy according to the task information and the comprehensive score value; packaging the task information into a communication data packet, sending the communication data packet to the optimal target node to execute the task, and generating a task result; returning a task result to the task initiating node to form a complete task processing closed loop; according to the method, the edge nodes negotiate through the lightweight protocol to complete task distribution and data transmission, so that the resource utilization rate of the edge nodes can be improved, the delay is reduced, and the stability and reliability of the system are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing, in particular to an edge-edge collaborative system and method supporting task perception and resource sharing. BACKGROUND

[0002] With the rapid growth of intelligent terminal devices (such as sensors, cameras, vehicle-mounted terminals, etc.), and the higher real-time processing requirements of artificial intelligence, high-definition video, automatic driving and other applications for computing resources, the traditional "edge-cloud" architecture gradually exposes its limitations. In high-concurrency and low-latency scenarios, relying on the central cloud for data processing and decision-making will increase communication delay and bandwidth pressure, and cannot meet the requirements of key businesses for response speed and service quality. Edge computing has thus become a new computing paradigm, which effectively alleviates the pressure on the cloud by deploying computing resources on edge nodes close to the terminal side, and improves the real-time performance and availability of the system.

[0003] Existing edge computing systems usually organize task processing procedures based on an "end-edge-cloud" integrated framework: terminal devices first complete preliminary perception and data collection locally, and then offload data to the nearest edge server through a wireless access network; edge nodes deploy inference services, data preprocessing and part of the business logic through virtualization technologies such as containers or virtual machines, to realize local computing of latency-sensitive tasks; for complex tasks with large computing volume or requiring a global perspective, they are further uploaded to the central cloud for centralized analysis and decision-making. Resource management and task scheduling rely on cloud-side controllers or pre-set static strategies, thereby achieving hierarchical collaboration and functional division among the end, edge and cloud to some extent.

[0004] However, the current edge computing architecture mainly focuses on vertical communication between "edge-cloud" or "end-edge-cloud", and lacks efficient collaborative mechanisms between horizontal "edge-edge". In actual applications, edge nodes often face problems such as limited computing resources and uneven task load. If there is a lack of task perception capability and reasonable collaborative scheduling strategy, it is easy to cause resource waste, service congestion and system bottlenecks. SUMMARY

[0005] In view of the lack of task perception capability and reasonable collaborative scheduling strategy in the prior art, which leads to resource waste, service congestion and system bottlenecks, the present application proposes an edge-edge collaborative system and method supporting task perception and resource sharing, based on real-time state perception and intelligent collaborative mechanisms among multiple edge nodes, to realize dynamic offloading of tasks and resource reuse, thereby solving the problems existing in the prior art.

[0006] An edge-edge collaborative system supporting task perception and resource sharing is deployed in multiple interconnected edge nodes with edge computing capability, which includes: A task perception module is configured to periodically collect resource state information of a task initiation node and generate a state vector in a unified format; and meanwhile, describe a latest arrived task of the node as structured task information; A resource evaluation module is connected to the task perception module; the resource evaluation module is configured to receive the state vector and calculate a comprehensive score value of each edge node for quantitatively representing the collaborative capability of the edge node; A collaborative scheduling module is connected to the task perception module and the resource evaluation module; the collaborative scheduling module is configured to receive the task information and the comprehensive score value, and dynamically select an optimal target node from the plurality of edge nodes based on a multi-factor scoring mechanism and a jitter prevention strategy; A communication management module is connected to the collaborative scheduling module; the communication management module is configured to encapsulate the task information as a communication data packet and send the communication data packet to the target node; A task execution module is connected to the collaborative scheduling module and the communication management module; the task execution module is configured to receive the target node and execute the task, generate a task result, and return the task result to the task initiation node.

[0007] Further, the resource evaluation module adopts a dual-track evaluation mechanism combining a rule engine and a lightweight learning model to calculate the comprehensive score value of each edge node, wherein the rule engine is configured to perform deterministic constraints and penalties, and the lightweight learning model is configured to adaptively calibrate the score weight according to historical execution feedback.

[0008] Further, the collaborative scheduling module internally organizes a scheduling process in a phased state machine, including an evaluation input, a candidate generation, a master-backup selection, a delivery and observation, a result back-feeding and a weight re-learning phase, and realizes fairness and isolation between tasks through a tenant quota and a weighted fair queue mechanism.

[0009] The application further provides an edge-to-edge collaborative method supporting task perception and resource sharing, which is collaboratively executed by a plurality of edge nodes and specifically includes the following steps: A plurality of edge nodes periodically collect resource state information of a task initiation node and generate a state vector in a unified format; and meanwhile, describe a latest arrived task of the node as structured task information; Based on the state vector, a comprehensive score value of each edge node for quantitatively representing the collaborative capability of the edge node is calculated; According to the task information and the comprehensive score value, an optimal target node is dynamically selected from the plurality of edge nodes based on a multi-factor scoring mechanism and a jitter prevention strategy; The task information is encapsulated as a communication data packet and sent to the optimal target node to execute the task, and a task result is generated; the task result is returned to the task initiation node to complete the task processing.

[0010] Further, by smoothing and standardizing the collected resource state information, a time-stamped state vector is encapsulated.

[0011] Further, based on the received state vector, a comprehensive score value of each edge node is calculated by combining a rule engine and a learning model evaluation algorithm; the comprehensive score value considers the remaining amount of computing resources, communication delay, link bandwidth and node health degree.

[0012] Further, the dynamic selection of the optimal target node from the plurality of edge nodes based on the multi-factor scoring mechanism and the anti-jitter strategy according to the task information and the comprehensive score value specifically includes the following steps: Based on the comprehensive score value, a candidate node set meeting the task execution feasibility condition is screened out; The candidate node set is sorted according to the high and low of the comprehensive score value; Based on the anti-jitter strategy, by setting a minimum score difference threshold and a minimum residence time, a final target node is selected from the sorted candidate node set.

[0013] Further, it also includes that, in the process of dynamically selecting the optimal target node from the plurality of edge nodes based on the multi-factor scoring mechanism and the anti-jitter strategy, for a critical task with a priority greater than a preset threshold, at least one backup node is selected in addition to the selected main target node; when the task distribution to the main target node fails or times out, the task distribution process to the backup node is automatically triggered.

[0014] The present application provides an edge-edge collaborative system supporting task awareness and resource sharing, which has the following beneficial effects: The present application deploys a task awareness module in each edge node, which is used to collect information such as the computing resource utilization rate of the task initiation node, the current task load, the task queue length and the network connection state in real time, and by constructing a node capability portrait, the system can comprehensively perceive the running state of each edge node; the running state information of each edge node is used as the basis for node collaboration, and the resource evaluation module performs dynamic scoring to evaluate the collaborative ability of the node, thereby providing a decision basis for subsequent task offloading; this process effectively avoids single node overload and computing resource waste, and improves the overall load balancing and task response efficiency of the system; based on task awareness, the present application proposes a collaborative scheduling strategy, which comprehensively models the key indicators of each edge node and dynamically selects the optimal task collaboration node accordingly, thereby realizing optimal offloading path decision of the task; thereby solving the problems of limited computing resources and uneven task load faced by edge nodes in practical applications; the system has good autonomy and scalability, can adapt to heterogeneous devices and complex network environments, and is widely applicable to high real-time scenarios such as industrial Internet of Things, intelligent transportation, edge intelligent security and the like. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an edge-to-edge collaborative system that supports task awareness and resource sharing in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] This invention proposes an edge-to-edge collaborative system supporting task awareness and resource sharing. Deployed across multiple edge nodes with edge computing capabilities, this system is suitable for scenarios such as rail transit environmental monitoring that require collaborative processing of track temperature and humidity, structural vibration, and video surveillance data. Based on real-time status awareness and intelligent collaboration mechanisms among multiple edge nodes, it enables dynamic task offloading and resource reuse. The system includes a task awareness module, a resource assessment module, a collaborative scheduling module, and a secure communication module. Each edge node negotiates through a lightweight protocol to complete task distribution and data transmission. This invention improves edge node resource utilization, reduces latency, and enhances system stability and reliability, making it particularly suitable for multi-source, task-intensive edge intelligent applications. An edge node refers to a physical or virtual computing unit deployed at the network edge, possessing independent computing, storage, and network communication capabilities.

[0018] like Figure 1 As shown, the system specifically includes: (1) Task awareness and status assessment phase between edge nodes. This phase is mainly responsible for sensing the resource status and task load of the nodes themselves, and performing quantitative assessment of resource capabilities accordingly, providing data support for the scheduling module; its corresponding modules and data flows are as follows:

[0019] ①Task Awareness Module → Resource Evaluation Module (State_Vector). State_Vector is used to describe the current resource status of edge nodes (such as CPU, memory, bandwidth, queue length, etc.).

[0020] Task awareness first triggers status collection of each node at a configurable period (default 1s, adjustable from 0.2-2s), covering CPU / memory / GPU memory usage, estimated available computing power (GFLOPs / TOPS), bandwidth, round-trip latency, packet loss, queuing time and number of tasks in transit, as well as battery power, GPU temperature and frequency, and key operator and image availability. To avoid amplifying decision noise from instantaneous fluctuations, the collected values ​​are first sampled at a granularity of 100-200ms, then subjected to exponential moving average and outlier pruning, followed by standardization and uniform encapsulation into a timestamped state vector, along with a health probability inferred from the stability window and alarm count. If individual indicators are missing, they are filled with the most recent valid value or historical P50 and the weight of that indicator in the scoring is automatically reduced; if the collection process is abnormal, the side vehicle monitoring automatically recovers and temporarily switches to "degraded reporting" (only retaining CPU, memory, bandwidth, and round-trip latency) to ensure continuous profile. Once completed, lightweight RPC or streaming channels will stably send the state vector into the resource assessment module, providing a consistent state foundation for subsequent decisions.

[0021] ②Task Awareness Module → CollaborativeScheduling Module (Task_Info), which contains basic descriptions of the task such as task type, computational requirements, and deadline.

[0022] As soon as the state base is ready, new tasks are profiled as structured task information objects, including key elements such as task identifier, type, estimated computation and memory requirements, data size and location, deadline and priority, security level, and tenant information. The estimated computation latency is based on the "relative relationship between the task's computational requirements and available computing power," while the estimated network latency is based on a "comprehensive assessment of the amount of data to be transmitted, available effective bandwidth, and path round-trip latency." These two factors are combined to obtain a baseline completion time. If a security level or data location is temporarily unavailable, the system uses a high-level protection and "local priority" as conservative defaults, and marks the task information as "incomplete," prompting the scheduler to adopt a more cautious strategy. Subsequently, this task profile, along with the continuously updated state flow from ①, enters the collaborative scheduling module, allowing scheduling to naturally transition from "knowing what to do" to "who is more suitable to complete it, and where."

[0023] ③ Resource Evaluation Module → Collaborative Scheduling Module (Score_Value): Score_Value represents the comprehensive score of a node's current collaborative capability and is used for scheduling decisions.

[0024] Once the task profile and node status are merged on the scheduling side, resource assessment provides a feasibility judgment and an interpretable comprehensive score for each candidate node. The scoring considers multiple factors: for example, "computing power availability" is understood as the inverse manifestation of CPU utilization; the remaining margin of memory and GPU memory is used as a hard threshold (below 10% is directly judged as infeasible); bandwidth, round-trip latency, and packet loss are combined into a network quality indicator; average waiting time is used to measure queue congestion; power consumption and battery life are combined to measure energy efficiency and endurance; the normalized value of node health probability and mean time between failures (MTBF) is combined to measure reliability; and data affinity is reflected by whether the data and task are located in the same domain or rack, while security domain matching is checked. If the baseline completion time is later than the deadline, or there is a mismatch in security domains, penalties are introduced for both deadlines and security; for nodes with high load hotspots or a high number of historical defaults, additional hotspot penalties are applied to suppress skew. The final score is obtained by "positive contribution minus various penalties," and includes the top three positive factors and the two main sources of penalties, facilitating more robust actions by the scheduler in the next step.

[0025] (2) Resource sharing and intelligent scheduling stage among edge nodes; this stage mainly realizes task offloading path selection, communication relay and remote execution scheduling to ensure resource sharing and intelligent scheduling in a multi-node collaborative environment. The corresponding modules and data flows are as follows:

[0026] ④ Collaborative Scheduling Module → Communication Management Module (Target_Node_ID), where Target_Node_ID is the edge node number to which the task should be unloaded, used to guide the communication module in issuing tasks.

[0027] Based on the sorting list in step ③, the collaborative scheduling first eliminates infeasible nodes (such as insufficient video memory, missing images, or compliance mismatches), and then selects the best from the compliant set: the node with the highest comprehensive score is selected as the primary choice; at the same time, a minimum score difference threshold and a minimum dwell time are set to avoid oscillations caused by frequent switching. For urgent or high-priority tasks, a small number of alternatives can be selected as hot backups, with health probe endpoints and validity periods attached, and the routing plan is then handed over to the communication management module. If no feasible external node is available at this time, the task will fall back to local execution, and the reason for failure will be written to the scheduling log as a basis for subsequent learning and weight rebalancing. Thus, through the continuous process of "screening—selection—bouncing—backup," the determination of the target node and the preparation for transmission are naturally connected.

[0028] ⑤ Collaborative Scheduling Module → Communication Management Module (Task_Info), which contains basic descriptions of the task such as task type, computational requirements, and deadline.

[0029] Once the objective is clear, the scheduler encapsulates the task information and routing plan into communication messages, which are then transmitted to the communication management module. To ensure end-to-end integrity and SLO verifiability, the messages include a model or mirror fingerprint, a list of data blocks and an integrity check tree, the maximum allowable latency and acceptable default rate, retry and timeout policies (the first timeout is typically set to approximately twice the round-trip latency, using exponential backoff), service level and priority, and are uniformly infused with timestamps and signatures or random numbers to prevent replay. Based on the link status, the communication side adaptively selects between QUIC and TCP; in weak network scenarios, forward error correction or multipathing can be enabled; for large-volume tasks, fixed-granularity fragmentation and parallel processing with verification and reassembly are used. If the main link handshake fails or consecutive latency anomalies occur, a smooth switch to a backup is initiated, and the scheduler is synchronously written back to ensure reliable task execution.

[0030] ⑥ Collaborative Scheduling Module → TaskExecution Module (Task Information): Task_Info contains basic descriptions of the task, such as task type, computational requirements, and deadline.

[0031] When the scheduler determines that local availability is better or external availability is unavailable, the execution module starts locally: first, it enqueues data based on priority and deadline, then reserves CPU / GPU / video memory and bandwidth through the quota manager; if the image is missing, it pre-fetches data in parallel and utilizes layer reuse to reduce cold starts; once the data is ready, it enters the running state. During execution, progress, estimated completion time, and SLO deviation are reported at 1-2 second intervals; if the deviation remains positive, it prioritizes increasing resources (increasing priority, releasing higher power limits, or increasing parallelism), and if this is still insufficient and the policy allows, it triggers parallel replicas for "firefighting," and after completion, it deduplicates the results and retains the optimal one. If the task fails or resources suddenly drop, it returns to step ④ to trigger rescheduling, and adds the current node to the time-limited blacklist to prevent bouncing. After the task is completed, a result object is generated and sent back to the scheduler and communication side, forming a traceable record and providing feedback for the next round of profile and weight updates.

[0032] ⑦ Communication Management Module → External Network (Communication Data Packet Comm_Packet): Comm_Packet encapsulates task information and target node structured data packets for inter-node transmission.

[0033] The communication management module encapsulates the target node identifier and task information into a communication data packet Comm_Packet, and offloads the task to the target edge node through the network to realize collaborative task transmission between nodes.

[0034] ⑧ Communication Management Module → Task Execution Module (Task Information Task_Info), Task_Info contains basic descriptions of the task such as task type, computation requirements, and deadline.

[0035] When the communication management module receives an unload task from another node, it will parse out the task information Task_Info and forward it to the local task execution module to trigger the execution of the collaborative task.

[0036] ⑨ Task Execution Module → Communication Management Module (Task_Result): Task_Result is the output data generated after the task is completed, which needs to be returned to the scheduling source or processing node.

[0037] After the task is completed, the task execution module generates the corresponding result data Task_Result and sends it to the communication management module, which decides whether to send it back to the original request node based on the task source.

[0038] To facilitate engineering implementation and review alignment, the key implementation units of the two mechanisms mentioned above will be explained together: The resource assessment module, as the core carrier of the "task awareness and status assessment mechanism," mainly runs through data flows ① and ③, responsible for transforming scattered node signals into interpretable and computable capability profiles and scoring inputs. The collaborative scheduling module, as the central node of the "resource sharing and intelligent scheduling mechanism," covers data flows ④-⑥ and, together with communication management and task execution, closes the transmission and feedback paths ⑦-⑨. The following sections will provide a comprehensive introduction to these two modules from three dimensions: data model, decision logic, and governance methods.

[0039] The resource assessment module is responsible for compressing noisy, heterogeneous, and time-varying node signals into a stable and interpretable "feasibility + score" that is friendly to scheduling. To this end, the module maintains a unified data model (State_Vector + metadata + healthy_prob), using a circular cache to hold the most recent N frame snapshots and asynchronously storing them in the time-series library as needed, ensuring that the past is visible and the present is clearly interpretable. On the other hand, it uses a dual-track approach of "rule engine + lightweight learning": the rule engine is responsible for deterministic constraints on thresholds, weights, and penalties, while the learning channel (such as GBDT / linear models) absorbs feedback from SLO results, actual ETC, and failure reasons, periodically calibrating each weight and hotspot penalty to form a scoring surface that adapts to load and environment. To reduce the cascading effects of observation errors, the module performs anomaly pruning, short-window smoothing, and missing data backfilling at the source, and attaches the top three contributors and main penalty sources to each score, achieving consistent and auditable "number-model-cause" consistency. When the center is unreachable, the score can be downgraded and buffered at the edge, and then batch-uploaded after recovery, ensuring that scheduling still has a "sufficiently good" profile available.

[0040] The collaborative scheduling module prioritizes compliance and SLO (Solution-Oriented Loop) performance. Beyond feasibility screening, it employs a stable path decision-making process comprised of "optimal ranking + jitter suppression + backup design." Internally, it organizes the process using a phased state machine: input evaluation → candidate generation → primary / backup selection → deployment and observation → result feedback and weight relearning. To balance efficiency and steady-state operation, the module introduces a difference threshold and minimum dwell time, avoiding frequent migrations when scores are stuck while allowing rapid switching during load and link changes. For high-priority or tight-deadline tasks, Top-K hot standby and idempotent deployment ensure seamless continuation in case of anomalies. Fairness and isolation in scheduling are achieved through tenant quotas and WFQ (Workforce-Free Queuing) rate limiting. In congested scenarios, low-priority tasks are gently degraded (e.g., reduced parallelism / delayed production). Simultaneously, the module fully links the input, interpretation, deployment, and implementation results of each decision to logs / metrics / trace, supporting SLO review, root cause analysis, and policy canary deployment and rollback. Through the above mechanisms, scheduling and resource assessment form a complementary, closed-loop, and evolving integrated capability.

[0041] This invention deploys a task-aware module in each edge node to collect real-time information such as the computing resource utilization of the task initiating node, the current task load, the task queue length, and the network connection status. By constructing a node capability profile, the system can comprehensively perceive the operating status of each edge node. This information serves as the basis for node collaboration, allowing the resource evaluation module to dynamically score the nodes and assess their collaborative capabilities, thus providing a decision-making basis for subsequent task unloading. This mechanism effectively avoids single-node overload and computing resource waste, improving the overall load balancing and task response efficiency of the system. Based on task awareness, this invention designs a collaborative scheduling strategy based on a multi-factor scoring mechanism. This strategy comprehensively models key indicators such as communication latency, link bandwidth, and remaining computing resources of each edge node, and dynamically selects the optimal task collaboration node accordingly, achieving optimal task unloading path decision-making. During collaboration, each edge node exchanges status and negotiates collaboration through a lightweight encrypted communication protocol, ensuring low latency and communication integrity. This system possesses good autonomy and scalability, can adapt to heterogeneous devices and complex network environments, and is widely applicable to high real-time scenarios such as industrial IoT, smart transportation, and edge intelligent security.

[0042] Based on the same inventive concept, this invention also proposes an edge-to-edge collaborative method that supports task awareness and resource sharing, which is executed collaboratively by multiple edge nodes and includes the following steps: Multiple edge nodes periodically collect resource status information from task initiation nodes. The collected resource status information is sampled at a granularity of 100-200ms and then subjected to exponential moving average, outlier pruning, and standardization to encapsulate it into a timestamped state vector. At the same time, newly arriving tasks are described as structured task information.

[0043] Based on state vectors, an evaluation algorithm combining a rule engine and a learning model is used to calculate a comprehensive score for each edge node to quantitatively represent its collaborative ability.

[0044] Based on task information and comprehensive score, the optimal target node is dynamically selected from multiple edge nodes using a multi-factor scoring mechanism and anti-jitter strategy.

[0045] The task information is encapsulated into a communication data packet and sent to the optimal target node to execute the task and generate the task result; the task result is then returned to the task initiating node to form a complete task processing closed loop.

[0046] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A side-to-side collaboration system supporting task awareness and resource sharing, characterized in that, The system is deployed in multiple interconnected edge nodes with edge computing capabilities, and the system comprises: A task perception module is configured to periodically collect resource state information of a task initiation node, and generate a state vector in a unified format; and the latest task description of the node is configured to be structured as structured task information; A resource evaluation module is connected with the task perception module; the resource evaluation module is configured to receive the state vector and calculate a comprehensive score value of each edge node for quantitatively representing the coordination capability of the edge node; A coordination scheduling module is connected with the task perception module and the resource evaluation module; the coordination scheduling module is configured to receive the task information and the comprehensive score value, and dynamically select an optimal target node from the multiple edge nodes based on a multi-factor scoring mechanism and a jitter prevention strategy; A communication management module is connected with the coordination scheduling module; the communication management module is configured to encapsulate the task information into a communication data packet and send the communication data packet to the target node; A task execution module is connected with the coordination scheduling module and the communication management module; the task execution module is configured to receive the target node and execute the task, generate a task result, and return the task result to the task initiation node.

2. The peer-to-peer system that supports task awareness and resource sharing according to claim 1, wherein, The resource evaluation module adopts a dual-track evaluation mechanism combining a rule engine and a lightweight learning model to calculate the comprehensive score value of each edge node, wherein the rule engine is configured to execute deterministic constraints and penalties, and the lightweight learning model is configured to adaptively calibrate the score weight according to historical execution feedback.

3. The peer-to-peer system that supports task awareness and resource sharing according to claim 1, wherein, The coordination scheduling module internally organizes a scheduling process in a phased state machine, including an evaluation input, a candidate generation, a master-slave selection, a delivery and observation, a result back-feeding and a weight re-learning phase, and realizes the fairness and isolation between tasks through a tenant quota and a weighted fair queue mechanism.

4. A method for edge-edge collaboration in support of task awareness and resource sharing, the method comprising: The method is cooperatively executed by multiple edge nodes, and specifically comprises the following steps: The multiple edge nodes periodically collect resource state information of a task initiation node, and generate a state vector in a unified format; and the latest task description of the node is configured to be structured as structured task information; Based on the state vector, a comprehensive score value of each edge node for quantitatively representing the coordination capability of the edge node is calculated; According to the task information and the comprehensive score value, an optimal target node is dynamically selected from the multiple edge nodes based on a multi-factor scoring mechanism and a jitter prevention strategy; The task information is encapsulated into a communication data packet and sent to the optimal target node to execute the task, and a task result is generated; the task result is returned to the task initiation node to complete the task processing.

5. The peer-to-peer collaboration method of supporting task awareness and resource sharing according to claim 4, wherein, The collected resource state information is sampled at a granularity of 100-200 ms, and then subjected to exponential moving average, abnormal value clipping processing and standardization to encapsulate a state vector with a time stamp.

6. The peer-to-peer collaboration method of supporting task awareness and resource sharing according to claim 4, wherein, Based on the received state vector, an evaluation algorithm combining a rule engine and a learning model is used to calculate a comprehensive score value of each edge node; the comprehensive score value considers the remaining amount of computing resources, communication delay, link bandwidth and node health degree.

7. The peer-to-peer collaboration method of supporting task awareness and resource sharing according to claim 4, wherein, The optimal target node is dynamically selected from the multiple edge nodes based on the task information and the comprehensive score value, and based on a multi-factor scoring mechanism and a jitter prevention strategy, specifically comprising the following steps: Based on the comprehensive score value, a candidate node set meeting the task execution feasibility condition is screened out; The candidate node set is sorted according to the comprehensive score value; Based on the anti-jitter strategy, a final target node is selected from the sorted candidate node set by setting a minimum score difference threshold and a minimum residence time.

8. The peer-to-peer collaboration method of supporting task awareness and resource sharing according to claim 4, wherein, Further comprising, in the process of dynamically selecting the optimal target node from multiple edge nodes based on the multi-factor scoring mechanism and the anti-jitter strategy, for a critical task with a priority greater than a preset threshold, in addition to the selected main target node, at least one backup node is also selected; when the task distribution to the main target node fails or times out, the task distribution process to the backup node is automatically triggered.

Images