Collaborative computing management method for hyperconverged systems with multi-level edge nodes on highways
By using heterogeneous resource pooling and multi-objective optimization offloading methods, the problems of uneven resource allocation and inflexible task offloading in highway edge computing management are solved, realizing dynamic scheduling of edge nodes and high system availability, and ensuring the continuity of business flow and processing efficiency.
Patent Information
- Application Number
- CN202611288355.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing highway edge computing management solutions suffer from problems such as uneven resource allocation, inflexible task offloading, and insufficient consideration of node health status when dealing with complex and ever-changing business loads and network environments, resulting in low system processing efficiency and poor business continuity.
This paper adopts a method that combines heterogeneous resource pooling, multi-objective optimization offloading, and offline autonomous control. By acquiring the heterogeneous computing resources, storage resources, and network bandwidth parameters of edge nodes, edge computing power resource pool parameters are generated, health measurement analysis is performed, a multi-level edge topology diagram is constructed, and graph neural networks and distributed game theory optimization algorithms are used to dynamically offload tasks, thereby achieving efficient scheduling of computing tasks and high system availability.
It achieves dynamic balanced scheduling of edge computing resources, ensuring high availability and business continuity of the system, and can automatically adjust strategies in the event of network anomalies to ensure the stable operation of critical services.
Smart Images

Figure CN123123124A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edge computing and distributed system resource management technology, and in particular to a collaborative computing management method for hyperconverged systems with multi-level edge nodes on highways. Background Technology
[0002] The advancement of highway information technology construction has led to a continuous expansion of the deployment scale of roadside equipment and sensors, which generate massive amounts of real-time data during operation. To meet the demands of vehicle-to-infrastructure (V2I) and intelligent monitoring for low latency and high concurrency processing, edge computing technology is widely used in highway scenarios. Multi-level edge nodes effectively alleviate bandwidth pressure and processing bottlenecks on central cloud servers by pushing computing resources down to the roadside closer to the data source. Hyperconverged systems further deeply integrate computing, storage, and network resources, providing edge nodes with a more flexible and scalable infrastructure support.
[0003] Existing highway edge computing management solutions typically employ static resource allocation strategies and fixed task offloading rules. When processing business flow data generated by roadside equipment, the system often relies on pre-set thresholds to trigger task migration, allocating computing tasks to nearby single edge nodes or directly transmitting them back to the central cloud server. For scheduling heterogeneous hardware resources, existing solutions mainly use simple virtualization technology to logically partition physical resources, lacking unified abstraction and pooling management of global computing power resources.
[0004] Existing technologies have limitations in dealing with the complex and ever-changing business loads and network environments of highways. Static resource allocation strategies struggle to adapt to the dynamic differences in latency and interaction intensity required by different business flows, easily leading to overloaded nodes while other nodes remain idle. Fixed task offloading rules do not adequately consider the real-time health status of edge nodes and the global network topology. When a node experiences a hidden fault or network connectivity is unstable, the scheduling strategy cannot be adjusted in a timely manner, thus affecting the continuity of critical services and the overall processing efficiency of the system. Summary of the Invention
[0005] To address the aforementioned issues, this invention provides a collaborative computing management method for hyperconverged systems of multi-level edge nodes on highways. It employs a combination of heterogeneous resource pooling, multi-objective optimization offloading, and offline autonomous control techniques to achieve dynamic balanced scheduling of edge computing resources and high-availability operation of the system.
[0006] The above objectives can be achieved through the following approach:
[0007] A collaborative computing management method for hyperconverged systems with multi-level edge nodes on highways, including:
[0008] Obtain heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing;
[0009] Obtain the operational indicator parameters of the multi-level edge nodes, perform health metric analysis on the operational indicator parameters, and generate a real-time availability index for the nodes.
[0010] Acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels;
[0011] Obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, construct a multi-level edge topology diagram by combining the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels.
[0012] The multi-level edge topology graph is input into the collaborative unloading model for multi-objective optimization processing to generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm.
[0013] According to the dynamic task unloading strategy, the target computing task is migrated to the target edge node for processing, task execution feedback parameters are generated, and the collaborative unloading model is adaptively corrected based on the task execution feedback parameters.
[0014] Optionally, the acquisition of heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of the highway, through resource abstraction processing, includes:
[0015] Obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of devices with different architectures in the multi-level edge nodes of the highway to obtain the underlying heterogeneous resource parameters;
[0016] The underlying heterogeneous resource parameters are containerized and encapsulated into lightweight virtual machines to generate standardized virtual resource parameters;
[0017] The standardized virtual resource parameters are aggregated across nodes to generate edge computing resource pool parameters.
[0018] Optionally, obtaining the operational indicator parameters of the multi-level edge nodes and performing health measurement analysis on the operational indicator parameters includes:
[0019] Obtain the hardware operating status data and environmental monitoring data of the multi-level edge nodes to obtain operating index parameters;
[0020] The operating index parameters are subjected to latent fault prediction feature extraction to generate a hardware degradation trend feature vector.
[0021] The real-time availability index of a node is generated by combining the hardware degradation trend feature vector with a preset utility evaluation function for multi-dimensional weighted calculation.
[0022] Optionally, the step of acquiring business flow data generated at the edge of the highway and performing multi-level classification processing on the business flow data includes:
[0023] Obtain raw traffic flow data generated at the edge of the highway, perform feature parsing on the raw traffic flow data, and generate traffic flow feature parameters;
[0024] Based on the latency requirement and interaction intensity features in the business flow characteristic parameters, the original business flow data is clustered to generate business flow clusters.
[0025] Priority mapping is performed on the business flow clusters to generate business flow classification labels.
[0026] Optionally, the local load parameters of the multi-level edge nodes are obtained. When the local load parameters exceed a preset load threshold, a multi-level edge topology diagram is constructed by combining the edge computing resource pool parameters, the node real-time availability index, and the service flow classification label, including:
[0027] Obtain the real-time computing request data of the multi-level edge nodes and calculate the local load parameters;
[0028] When the local load parameter exceeds the preset load threshold, the edge computing power resource pool parameter and the node real-time availability index are extracted as the node feature matrix.
[0029] By combining the node feature matrix, the service flow classification labels, and the network connection relationships between nodes, a multi-level edge topology diagram is constructed.
[0030] Optionally, inputting the multi-level edge topology diagram into the collaborative unloading model for multi-objective optimization includes:
[0031] The multi-level edge topology graph is input into a graph neural network for graph node representation learning to generate a global network state feature vector.
[0032] The global network state feature vector is input into a distributed game theory optimization algorithm, and the algorithm iteratively solves the problem with the goal of minimizing processing latency and maximizing resource balance to generate a task allocation matrix.
[0033] The task allocation matrix is parsed to generate a dynamic task unloading strategy.
[0034] Optionally, according to the dynamic task unloading strategy, migrating the target computing task to the target edge node for processing, generating task execution feedback parameters, and adaptively correcting the collaborative unloading model based on the task execution feedback parameters include:
[0035] According to the dynamic task unloading strategy, the target computing task is scheduled to the target edge node for processing;
[0036] Obtain the actual latency and energy consumption parameters of the target edge node after executing the target computing task, and generate task execution feedback parameters;
[0037] A gradient update function is constructed based on the task execution feedback parameters to adaptively adjust the network weight parameters of the graph neural network and the distributed game theory optimization algorithm.
[0038] Optionally, the method further includes:
[0039] During the dynamic unloading and execution of the task, the network connection status parameters between the multi-level edge nodes and the upper-level central node are obtained;
[0040] When the network connection status parameter indicates a network interruption, the local policy database of the edge node is activated to generate offline autonomous control instructions;
[0041] According to the offline autonomous control instructions, the operation of local key business processes is controlled, and the business flow data is compressed and stored to generate local incremental cache data.
[0042] Optionally, the method further includes:
[0043] The network connection status parameters are continuously monitored, and when network recovery is detected, the local incremental cache data is obtained.
[0044] The local incremental cache data is timestamped and prioritized to generate a breakpoint resume data stream;
[0045] The interrupted data stream is smoothly and adaptively adjusted based on the current network bandwidth parameters and synchronized to the upper-level central node.
[0046] Based on the same inventive concept, this invention also provides a hyper-converged system collaborative computing management system for multi-level edge nodes of highways, the system comprising:
[0047] The resource abstraction module is used to obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing.
[0048] The health analysis module is used to obtain the operation index parameters of the multi-level edge nodes, perform health quantification analysis on the operation index parameters, and generate a real-time availability index for the nodes.
[0049] The business classification module is used to acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels.
[0050] The topology construction module is used to obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, the module combines the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels to construct a multi-level edge topology structure diagram.
[0051] The collaborative optimization module is used to input the multi-level edge topology graph into the collaborative unloading model for multi-objective optimization processing and generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm.
[0052] The adaptive correction module is used to migrate the target computing task to the target edge node for processing according to the task dynamic unloading strategy, generate task execution feedback parameters, and adaptively correct the collaborative unloading model based on the task execution feedback parameters.
[0053] Compared with the prior art, the present invention has the following advantages:
[0054] This invention constructs a unified edge computing resource pool by standardizing container encapsulation and cross-node pooling aggregation of heterogeneous computing, storage, and network resources of multi-level edge nodes. At the same time, it combines the hardware operating status of the devices and environmental monitoring data to perform latent fault prediction and multi-dimensional weighted health metric analysis, generating a real-time availability index for nodes, thereby realizing the overall dynamic management of global computing resources and ensuring high system availability.
[0055] This invention generates business flow classification labels by performing feature analysis, clustering, and priority mapping on the original business flows generated at the edge of highways. When the local load exceeds the limit, it constructs a multi-level edge topology diagram by combining edge computing resource pool parameters, node availability index, and business flow labels. It then inputs a collaborative offloading model containing graph neural networks and distributed game theory optimization algorithms for multi-objective optimization, thereby achieving efficient dynamic scheduling of computing tasks in complex network environments.
[0056] This invention schedules computational tasks to target edge nodes for execution based on a dynamic task offloading strategy, obtains actual latency and energy consumption feedback parameters, constructs a gradient update function to iteratively correct the network weight parameters of the model, and automatically activates the local policy database to ensure the operation of critical services and incremental data compression caching when the network is interrupted. After the network is restored, it performs breakpoint resume based on timestamp verification and smooth adaptive adjustment, thus realizing adaptive correction of the model and ensuring business continuity under abnormal network conditions.
[0057] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims, and drawings. Attached Figure Description
[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0059] Figure 1 This is a flowchart illustrating the collaborative computing management method for multi-level edge nodes of highways in a hyper-converged system according to an embodiment of the present invention.
[0060] Figure 2 This is a line graph showing the health and availability index of multi-level edge nodes in an embodiment of the present invention.
[0061] Figure 3 This is a heatmap of service flow classification features and priority mapping in an embodiment of the present invention.
[0062] Figure 4 This is a schematic diagram of the structure of the hyperconverged system collaborative computing management system for multi-level edge nodes of highways according to an embodiment of the present invention. Detailed Implementation
[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0064] Reference Figure 1One embodiment of the present invention proposes a collaborative computing management method for hyperconverged systems of multi-level edge nodes of highways. It adopts a combination of heterogeneous resource pooling, multi-objective optimization offloading and offline autonomous control to achieve dynamic balanced scheduling of edge computing resources and high availability operation of the system.
[0065] The method described in this embodiment specifically includes:
[0066] Obtain heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing;
[0067] Obtain the operational indicator parameters of the multi-level edge nodes, perform health metric analysis on the operational indicator parameters, and generate a real-time availability index for the nodes.
[0068] Acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels;
[0069] Obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, construct a multi-level edge topology diagram by combining the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels.
[0070] The multi-level edge topology graph is input into the collaborative unloading model for multi-objective optimization processing to generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm.
[0071] According to the dynamic task unloading strategy, the target computing task is migrated to the target edge node for processing, task execution feedback parameters are generated, and the collaborative unloading model is adaptively corrected based on the task execution feedback parameters.
[0072] Optionally, the acquisition of heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of the highway, through resource abstraction processing, includes:
[0073] Obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of devices with different architectures in the multi-level edge nodes of the highway to obtain the underlying heterogeneous resource parameters;
[0074] The underlying heterogeneous resource parameters are containerized and encapsulated into lightweight virtual machines to generate standardized virtual resource parameters;
[0075] The standardized virtual resource parameters are aggregated across nodes to generate edge computing resource pool parameters.
[0076] Specifically, the system connects to devices with different architectures across multiple layers of the highway's edge nodes. These devices include roadside computing units and tollbooth servers. It reads data from the devices' built-in hardware monitoring modules to obtain heterogeneous computing resource parameters. These parameters represent the number of floating-point operations per second (Floating-Point Operations Per Second) of the device's central processing unit. The system also synchronously reads the remaining available space capacity of the hard disk controller to obtain storage resource parameters. Network bandwidth parameters are determined by sending test data packets to adjacent nodes and recording the transmission time; these represent the maximum amount of data that the network link can transmit per unit time. The heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters are then numerically integrated to obtain the underlying heterogeneous resource parameters. These underlying heterogeneous resource parameters reflect the total original available resources of the physical hardware. The integration process is as follows:
[0077] ;
[0078] in, It represents the underlying heterogeneous resource parameters and is dimensionless. This represents parameters of heterogeneous computing resources, measured in floating-point operations per second. This represents storage resource parameters, measured in gigabytes. This represents network bandwidth, measured in megabits per second. It represents the normalization coefficient of computing resources, with the dimension of number of floating-point operations per second, and is set based on the computing weight determined by actual measurement data of highway roadside equipment. This represents the normalization coefficient of storage resources, measured in gigabytes, and is based on storage weights determined from measured data of highway roadside equipment. This represents the bandwidth resource normalization coefficient, measured in megabits per second, and is based on bandwidth weights determined from measured data of highway roadside equipment. The underlying heterogeneous resource parameters are containerized and encapsulated using lightweight virtual machines. Containerization and lightweight virtual machine encapsulation refer to using software isolation technology to divide physical hardware resources into independent and uniformly sized logical execution units. After this processing, the system generates standardized virtual resource parameters. These standardized virtual resource parameters represent the number of standard resource blocks that can be directly accessed by upper-layer services after isolation and partitioning. The calculation formula is:
[0079] ;
[0080] in, Represents standardized virtual resource parameters, dimensionless. It represents the resource requirements of the basic business unit, is dimensionless, and is set based on the basic business operation baseline determined by measured data of highway roadside equipment. This represents the encapsulation loss compensation factor, dimensionless, and is set based on the virtualization overhead ratio determined from measured data of highway roadside equipment. Standardized virtual resource parameters undergo cross-node pooling aggregation. Cross-node pooling aggregation refers to the logical merging and unified addressing of virtual resource blocks from nodes at different physical locations using a distributed coordination protocol. After merging, the system generates edge computing resource pool parameters. These parameters represent the total scale of globally schedulable virtual resources across the entire highway segment. The calculation formula is:
[0081] ;
[0082] in, This represents the parameters of the edge computing resource pool and is dimensionless. Represents the total number of nodes participating in the aggregation; it is dimensionless. Representing the Standardized virtual resource parameters for each node, dimensionless. Represents the network topology weight of nodes, is dimensionless, and is defined based on the node centrality index determined by measured data from highway roadside equipment.
[0083] For example, by accessing different architecture devices in a multi-level edge node of a highway, data from the hardware monitoring module is read to obtain heterogeneous computing resource parameters of 2000 floating-point operations per second. Data from the hard disk controller is read to obtain storage resource parameters of 500 gigabytes per second. Test data packets are sent to obtain network bandwidth parameters of 1000 megabits per second. Based on actual measurement data from highway roadside equipment, the normalization coefficients for computing resources are set to 0.01 floating-point operations per second, storage resources to 0.05 gigabytes per second, and bandwidth resources to 0.02 megabits per second. The system calculates the underlying heterogeneous resource parameters as follows: The system containerizes and encapsulates the underlying heterogeneous resource parameters using lightweight virtual machines. Based on measured data from highway roadside equipment, the resource requirement for the basic business unit is set to 5, and the encapsulation loss compensation factor is set to 0.9. The system calculates standardized virtual resource parameters: The system performs cross-node pooling aggregation processing on standardized virtual resource parameters. Assume there are 3 nodes, each with a standardized virtual resource parameter of 11.7. Based on measured data from highway roadside equipment, the node network topology weights are set to 1.0, 0.8, and 0.9, respectively. The system calculates the edge computing resource pool parameters: .
[0084] Optionally, obtaining the operational indicator parameters of the multi-level edge nodes and performing health measurement analysis on the operational indicator parameters includes:
[0085] Obtain the hardware operating status data and environmental monitoring data of the multi-level edge nodes to obtain operating index parameters;
[0086] The operating index parameters are subjected to latent fault prediction feature extraction to generate a hardware degradation trend feature vector.
[0087] The real-time availability index of a node is generated by combining the hardware degradation trend feature vector with a preset utility evaluation function for multi-dimensional weighted calculation.
[0088] Specifically, the system acquires real-time physical parameters of core components within multi-level edge nodes through a built-in sensor network, obtaining hardware operating status data. This hardware operating status data represents the real-time operating temperature of the device's central processing unit and motherboard. Simultaneously, external meteorological sensing devices collect temperature and humidity data of the physical space where the node is located, obtaining environmental monitoring data. This environmental monitoring data represents the comprehensive temperature equivalent of the external environment. The hardware operating status data and environmental monitoring data are numerically aligned and weighted to obtain operating indicator parameters. These operating indicator parameters reflect the overall operating pressure state of the device under the combined influence of the current internal and external environments. This fusion process is as follows:
[0089] ;
[0090] in, These represent operational parameters and are dimensionless. This represents hardware operating status data, measured in degrees Celsius. This represents environmental monitoring data, measured in degrees Celsius. The dimensionless coefficient represents the hardware status, with the dimension per degree Celsius. It is set based on the weight of internal temperature influence determined by pressure testing at highway edge nodes. The dimensionless coefficient representing environmental conditions, measured in degrees Celsius, is set based on the weights of external environmental influences determined by stress tests at highway edge nodes. Latent fault prediction features are extracted from operational index parameters. Latent fault prediction feature extraction refers to capturing the micro-trends of equipment performance degradation by comparing the rate of change of operational index parameters within adjacent time periods. This extraction process generates a hardware degradation trend feature vector. The hardware degradation trend feature vector represents the quantified probability of the equipment experiencing performance bottlenecks or downtime within a specific future time window. The calculation formula is:
[0091] ;
[0092] in, It represents the feature vector of hardware degradation trend and is dimensionless. The dimensionless parameter represents the operational performance at the current moment. The dimensionless parameter represents the operational performance parameters at the previous sampling time. This represents the sampling time interval, measured in seconds. The time compensation coefficient, measured in seconds, is set based on a time smoothing benchmark determined by stress testing of highway edge nodes. A multi-dimensional weighted calculation is performed, combining the hardware degradation trend feature vector with a preset utility evaluation function. The preset utility evaluation function refers to the basic health score of the equipment under ideal conditions, calibrated at the system's factory. The multi-dimensional weighted calculation refers to the process of comprehensively adjusting the basic health score and dynamic degradation risk. After calculation, the system generates a node real-time availability index. The node real-time availability index represents the comprehensive capability rating of the edge node's ability to stably handle new computing tasks at the current moment. The calculation formula is:
[0093] ;
[0094] in, The real-time availability index represents the number of nodes and is dimensionless. The output value of the preset utility evaluation function is dimensionless and is set based on the ideal health baseline determined by stress testing at highway edge nodes. The weight representing the attenuation penalty is dimensionless and is set based on the risk tolerance index determined by stress testing at highway edge nodes.
[0095] For example, the hardware operating status data of multi-level edge nodes is collected at 65 degrees Celsius. The environmental monitoring data is collected at 35 degrees Celsius. Based on the stress test of highway edge nodes, the hardware status simplification factor is set to 0.8 per degree Celsius, and the environmental status simplification factor is set to 0.2 per degree Celsius. The system calculates the following operating index parameters: 9. Assume the operating index parameter at the previous sampling time is 55, and the sampling time interval is 10 seconds. Based on the stress test of highway edge nodes, the time compensation coefficient is set to 5 seconds. The system extracts latent fault prediction features from the operating index parameters and calculates the hardware degradation trend feature vector: ( Based on stress tests at highway edge nodes, the preset utility evaluation function output value is set to 100, and the decay penalty weight is set to 3. The system combines the hardware decay trend feature vector with the preset utility evaluation function to perform multi-dimensional weighted calculations: Successfully generated the node real-time availability index. (Example:) Figure 2 As shown, the curves illustrating the evolution of the real-time availability index of different edge nodes during continuous operation, along with the hardware degradation trend, are illustrated. It can be seen that the availability index of each node exhibits a differentiated decline process based on its hardware degradation trend characteristics, providing an accurate quantitative basis for the subsequent construction of the topology graph and collaborative offloading.
[0096] Optionally, the step of acquiring business flow data generated at the edge of the highway and performing multi-level classification processing on the business flow data includes:
[0097] Obtain raw traffic flow data generated at the edge of the highway, perform feature parsing on the raw traffic flow data, and generate traffic flow feature parameters;
[0098] Based on the latency requirement and interaction intensity features in the business flow characteristic parameters, the original business flow data is clustered to generate business flow clusters.
[0099] Priority mapping is performed on the business flow clusters to generate business flow classification labels.
[0100] Specifically, raw traffic flow data generated at the edge of the highway is captured through network listening ports. This raw traffic flow data represents a collection of unprocessed data packets uploaded in real time by roadside cameras and sensors. Feature analysis is performed on the raw traffic flow data to extract packet header information and payload size, generating traffic flow feature parameters. These parameters reflect the core performance requirements of the data flow during transmission. These parameters are jointly determined by latency requirements and interaction strength characteristics. Latency requirements represent the maximum allowable data transmission and processing delay time for the business logic. Interaction strength characteristics represent the number of round trips between the device and the edge node per unit time. The traffic flow feature parameters are calculated as follows:
[0101] ;
[0102] in, These represent the characteristic parameters of the business flow and are dimensionless. It represents the latency requirement, and its dimension is milliseconds. It represents the intensity of interaction, measured in seconds per second. The normalized coefficient for latency is expressed in milliseconds and is set based on latency sensitivity weights determined through actual measurements of highway vehicle-road cooperative services. This represents the interaction normalization coefficient, measured in seconds per interaction, and is set based on the concurrent communication weight determined through actual measurements of highway vehicle-road cooperative services. The system clusters the original service flow data according to the latency requirements and interaction intensity characteristics in the service flow feature parameters. Clustering refers to grouping data flows with similar performance requirements into the same logical set. After the partitioning process, the system generates service flow clusters. Each service flow cluster represents a group identifier value for data flows with the same quality of service requirements. The calculation formula is:
[0103] ;
[0104] in, The representative business flow cluster is dimensionless. Represents the baseline value of basic characteristics, dimensionless, and is set based on the average value of conventional business characteristics determined by actual measurements of highway vehicle-road cooperative business. This represents a clustering adjustment factor, dimensionless, set based on the inter-cluster distance scaling ratio determined by actual measurements in highway vehicle-road cooperative services. Priority mapping is performed on the service flow clusters. Priority mapping converts cluster values into execution order levels directly recognizable by the system scheduler. After mapping, the system generates service flow classification labels. These labels represent the absolute priority ranking of the data flow in the edge node resource allocation queue. The calculation formula is:
[0105] ;
[0106] in, Represents business flow category labels, dimensionless. The priority mapping coefficient is dimensionless and is pre-set based on the level span ratio determined by actual measurements of highway vehicle-road cooperative services. It represents the basic priority offset, is dimensionless, and is set based on the minimum guarantee level value determined by actual measurements of highway vehicle-road cooperative services.
[0107] For example, the system captures raw traffic flow data generated at the edge of the highway, and parses it to obtain a latency requirement of 20 milliseconds and an interaction intensity of 50 interactions per second. Based on actual measurements of highway vehicle-road cooperative services, the latency normalization factor is set to 0.5 per millisecond, and the interaction normalization factor is set to 0.2 per interaction. The system calculates the traffic flow characteristic parameters: Based on actual measurements of vehicle-road cooperative services on highways, the baseline value for basic features is set to 10, and the clustering adjustment factor is set to 2. The system performs clustering on the original service flow data and calculates the service flow clusters: Based on actual measurements of highway vehicle-road cooperative services, the priority mapping coefficient is set to 1.5, and the basic priority offset is set to 1. The system performs priority mapping processing on service flow clusters and calculates service flow classification labels: Successfully generated business flow category tags.
[0108] Optionally, the local load parameters of the multi-level edge nodes are obtained. When the local load parameters exceed a preset load threshold, a multi-level edge topology diagram is constructed by combining the edge computing resource pool parameters, the node real-time availability index, and the service flow classification label, including:
[0109] Obtain the real-time computing request data of the multi-level edge nodes and calculate the local load parameters;
[0110] When the local load parameter exceeds the preset load threshold, the edge computing power resource pool parameter and the node real-time availability index are extracted as the node feature matrix.
[0111] By combining the node feature matrix, the service flow classification labels, and the network connection relationships between nodes, a multi-level edge topology diagram is constructed.
[0112] Specifically, the task scheduling interface listens to task instructions received by multi-level edge nodes to obtain real-time computing request data. This real-time computing request data represents the total number of computing tasks the node needs to process at the current moment. The real-time computing request data is compared with the node's own processing capacity to calculate local load parameters. These local load parameters reflect the current utilization rate of the node's computing resources. The calculation formula is:
[0113] ;
[0114] in, This represents local load parameters and is dimensionless. This represents real-time computation request data, measured in floating-point operations per second. It represents the node's processing capacity, measured in floating-point operations per second, and is a hardware benchmark computing power setting determined based on measured data from highway roadside equipment. This represents the load adjustment coefficient, dimensionless, set based on the concurrent task loss ratio determined by measured data from highway roadside equipment. Local load parameters are continuously monitored; when these parameters exceed a preset load threshold, a topology construction mechanism is triggered. The preset load threshold represents the maximum resource utilization ratio at which a node can operate stably. The system extracts edge computing resource pool parameters and the node's real-time availability index, concatenating them dimensionally to form a node feature matrix. This node feature matrix represents a comprehensive set of attributes integrating global computing power scale and node health status. Calculation formula:
[0115] ;
[0116] in, The matrix represents the feature matrix of the nodes and is dimensionless. This represents the parameters of the edge computing resource pool and is dimensionless. The real-time availability index represents the number of nodes and is dimensionless. The scaling factor represents the matrix, is dimensionless, and is set based on the alignment ratio of feature values determined by measured data from highway roadside equipment. A multi-level edge topology diagram is constructed by combining node feature matrices, service flow classification labels, and network connectivity relationships between nodes. Network connectivity relationships represent the connectivity status and communication quality of physical links between nodes. The multi-level edge topology diagram represents a network graph model reflecting global node resources, status, and service priorities. Calculation formula:
[0117] ;
[0118] in, It represents a multi-level edge topology diagram and is dimensionless. Represents business flow category labels, dimensionless. It represents network connectivity, is dimensionless, and is set based on the quantitative value of link connectivity determined by measured data from highway roadside equipment.
[0119] For example, the real-time computing request data for multi-level edge nodes is 8000 floating-point operations per second. Based on measured data from highway roadside equipment, the node processing capacity is set to 10000 floating-point operations per second, and the load adjustment coefficient is set to 1.1. The system calculates local load parameters: Assuming a preset load threshold of 0.8, the current local load parameter exceeds the preset load threshold. The system extracts an edge computing resource pool parameter of 31.59, and the node real-time availability index is 94. Based on measured data from highway roadside equipment, the matrix scaling factor is set to 0.01. The system calculates the node feature matrix: The system obtains a service flow classification label of 7. Based on measured data from highway roadside equipment, the network connectivity is set to 1.2. The system constructs a multi-level edge topology diagram: A multi-level edge topology diagram was successfully constructed.
[0120] Optionally, inputting the multi-level edge topology diagram into the collaborative unloading model for multi-objective optimization includes:
[0121] The multi-level edge topology graph is input into a graph neural network for graph node representation learning to generate a global network state feature vector.
[0122] The global network state feature vector is input into a distributed game theory optimization algorithm, and the algorithm iteratively solves the problem with the goal of minimizing processing latency and maximizing resource balance to generate a task allocation matrix.
[0123] The task allocation matrix is parsed to generate a dynamic task unloading strategy.
[0124] Specifically, a multi-level edge topology graph is input into the collaborative offloading model. The collaborative offloading model includes a graph neural network module and a distributed game theory optimization algorithm module. The graph neural network module is used to learn graph node representations of the multi-level edge topology graph. Graph node representation learning refers to extracting deep association features between nodes and their neighboring nodes through a multi-layered information transmission mechanism. After learning, the system generates a global network state feature vector. The global network state feature vector represents a dimensionality-reduced mathematical expression of the resource distribution and load pressure of the entire edge computing network at the current moment. Calculation formula:
[0125] ;
[0126] in, It represents the global network state feature vector and is dimensionless. It represents a multi-level edge topology diagram and is dimensionless. The weight matrix represents the graph neural network weight matrix, which is dimensionless and is based on the feature extraction weight settings determined during training of the highway collaborative unloading model. The bias term of the graph neural network is dimensionless and is set based on the baseline offset determined during training of the highway cooperative unloading model. The global network state feature vector is input into a distributed game theory optimization algorithm. Distributed game theory optimization algorithms are mathematical methods that find the optimal solution to a system by simulating the strategic interactions between multiple decision-makers. The algorithm iteratively solves the problem with the objectives of minimizing processing latency and maximizing resource balance. Iterative solving refers to the process of continuously adjusting the strategy in multiple rounds of computation until the system's payoff no longer increases significantly. After solving, the system generates a task allocation matrix. The task allocation matrix represents the set of mapping relationships between each computational task and the target edge node. Calculation formula:
[0127] ;
[0128] in, This represents the task allocation matrix, which is dimensionless. It represents the global network state feature vector and is dimensionless. The value represents the comprehensive cost function, which is dimensionless and is set based on the joint penalty term for time delay and energy consumption determined during the training of the highway cooperative unloading model. The convergence coefficients represent the game's convergence coefficients, are dimensionless, and are the iterative step size control parameters determined during training of the highway collaborative unloading model. The task allocation matrix undergoes policy parsing. Policy parsing refers to converting a mathematical matrix into a sequence of specific scheduling instructions executable by the system. After parsing, the system generates a dynamic task unloading policy. The dynamic task unloading policy represents the complete operational procedure guiding the migration of specific computational tasks from the source node to the target node. The calculation formula is:
[0129] ;
[0130] in, This represents a dynamic unloading strategy for tasks, and is dimensionless. This represents the task allocation matrix, which is dimensionless. The representative strategy conversion coefficient is dimensionless and is preset based on performance testing experience in actual business scenarios. This represents the basic strategy template value, which is dimensionless and is preset based on performance testing experience in actual business scenarios.
[0131] For example, a multi-level edge topology graph is input into a graph neural network, and the quantized representation value of the multi-level edge topology graph is 249.43464. Based on the training of the highway cooperative unloading model, the weight matrix of the graph neural network is set to 0.5, and the bias term of the graph neural network is set to 10. The system calculates the global network state feature vector: The global network state feature vector is input into the distributed game theory optimization algorithm. Based on the highway cooperative unloading model training, the comprehensive cost function value is set to 20, and the game convergence coefficient is set to 0.8. The system iteratively solves the problem and calculates the task allocation matrix: Based on the training of the highway collaborative unloading model, the policy transformation coefficient is set to 2, and the basic policy template value is set to 5. The system performs policy parsing on the task allocation matrix and calculates the dynamic unloading policy for the tasks: Successfully generated a dynamic uninstallation strategy for the task.
[0132] Optionally, according to the dynamic task unloading strategy, migrating the target computing task to the target edge node for processing, generating task execution feedback parameters, and adaptively correcting the collaborative unloading model based on the task execution feedback parameters include:
[0133] According to the dynamic task unloading strategy, the target computing task is scheduled to the target edge node for processing;
[0134] Obtain the actual latency and energy consumption parameters of the target edge node after executing the target computing task, and generate task execution feedback parameters;
[0135] A gradient update function is constructed based on the task execution feedback parameters to adaptively adjust the network weight parameters of the graph neural network and the distributed game theory optimization algorithm.
[0136] Specifically, based on the dynamic task offloading strategy, the target computing task is scheduled to the target edge node for processing via the underlying communication interface. The target computing task represents the specific business logic code and data set to be processed that require computing power offloading. The target edge node represents the physical hardware device that undertakes the computing task after allocation by the collaborative offloading model. After the target edge node completes the computation and returns the result, the actual latency and energy consumption parameters are obtained through the hardware monitoring module. The actual latency parameter represents the actual physical time consumed from the start of task execution to the output result. The energy consumption parameter represents the total electrical energy consumed by the target edge node during the processing of the task. The actual latency and energy consumption parameters are numerically aligned and weighted to generate task execution feedback parameters. The task execution feedback parameters represent the comprehensive performance evaluation quantification index of this task scheduling. This fusion process is as follows:
[0137] ;
[0138] in, This represents the task execution feedback parameters, which are dimensionless. This represents the actual time delay parameter, measured in milliseconds. This represents an energy consumption parameter, measured in joules. The time delay feedback coefficient is represented by the unit of milliseconds and is set based on the time cost weight determined by actual measurements of highway edge computing tasks. The energy consumption feedback coefficient, measured in joules, is set based on the electricity cost weight determined by actual measurements from highway edge computing tasks. A gradient update function is constructed based on the task execution feedback parameters. The gradient update function represents the mathematical rules used to calculate model decision errors and guide parameter adjustment. The gradient update function is used to adaptively correct the network weight parameters of graph neural networks and distributed game theory optimization algorithms. The network weight parameters represent the basic product coefficients used for feature extraction and policy generation within the collaborative unloading model. Adaptive correction refers to the process of automatically adjusting the model's internal parameters based on actual execution results to optimize future decisions. After correction, the system completes the closed-loop iteration of the model. The calculation formula is:
[0139] ;
[0140] in, This represents the updated network weight parameters, which are dimensionless. This represents the network weight parameters before the update, and is dimensionless. The learning rate is dimensionless and represents the parameter update step size set based on actual measurements of highway edge computing tasks. The expected feedback baseline value is dimensionless and is set based on the ideal scheduling effect index determined by actual measurements of highway edge computing tasks. This represents the gradient scaling factor, which is dimensionless and is set based on the error amplification ratio determined by actual measurements in highway edge computing tasks.
[0141] For example, the target computing task is scheduled to the target edge node for processing according to the dynamic task offloading strategy. The actual latency parameter obtained after the target edge node executes the target computing task is 40 milliseconds, and the energy consumption parameter is 15 joules. Based on the actual measurement of the highway edge computing task, the latency feedback coefficient is set to 0.5 per millisecond, and the energy consumption feedback coefficient is set to 0.8 per joule. Computation task execution feedback parameters: A gradient update function is constructed based on the task execution feedback parameters to adaptively adjust the network weight parameters. Assume the network weight parameters before the update are 0.5. Based on actual tests of a highway edge computing task, the learning rate is set to 0.01, the expected feedback baseline value is set to 20, and the gradient scaling factor is set to 0.5. The updated network weight parameters are calculated as follows: .
[0142] Optionally, the method further includes:
[0143] During the dynamic unloading and execution of the task, the network connection status parameters between the multi-level edge nodes and the upper-level central node are obtained;
[0144] When the network connection status parameter indicates a network interruption, the local policy database of the edge node is activated to generate offline autonomous control instructions;
[0145] According to the offline autonomous control instructions, the operation of local key business processes is controlled, and the business flow data is compressed and stored to generate local incremental cache data.
[0146] Specifically, during the dynamic unloading and execution of tasks, network connection status parameters between multi-level edge nodes and the upper-level central node are obtained by periodically sending heartbeat detection data packets. These network connection status parameters represent the connectivity and response latency of the communication link between the node and the central node. This parameter is quantified by comparing the actual response time with a preset timeout threshold. The calculation formula is as follows:
[0147] ;
[0148] in, Represents network connection status parameters, dimensionless. This represents the heartbeat response time, measured in milliseconds. This represents a preset timeout threshold, measured in milliseconds, based on the maximum tolerable latency setting determined through actual measurements in highway wide area network communication. Network connection status parameters are continuously monitored; when these parameters exceed a preset interruption threshold, a network interruption is identified. At this point, the local policy database of the edge node is activated. The local policy database represents a set of emergency handling rules pre-issued and stored in the node's local memory. Combining the basic instruction values from the local policy database with service flow classification tags, offline autonomous control instructions are generated. These instructions guide the node to independently allocate resources and degrade services during network outages. The formula is:
[0149] ;
[0150] in, Represents offline autonomous control commands; dimensionless. Represents the basic instruction value of the local policy database, dimensionless, and is based on the emergency response baseline set determined by actual measurements of highway wide area network communication. Represents business flow category labels, dimensionless. This represents the autonomous adjustment coefficient, dimensionless, set based on the offline degradation ratio determined by actual measurements in highway wide area network communication. According to offline autonomous control instructions, it controls the operation of local critical services, ensuring uninterrupted core functions. During this period, the system collects and generates business transaction data. This data represents business processing records and perception data generated during offline periods that would otherwise need to be uploaded to the central node in real time. The system uses a lossless compression algorithm to compress and store the business transaction data, generating local incremental cache data. This local incremental cache data represents data blocks that, after volume compression, are temporarily stored on the node's local hard drive awaiting synchronization after network recovery. The calculation formula is:
[0151] ;
[0152] in, This represents locally incremental cached data, measured in megabytes. This represents business transaction data, measured in megabytes. Represents the compression ratio, dimensionless, and is set based on the data compression limit ratio determined by actual measurements of highway wide area network communication. Represents the storage overhead coefficient, dimensionless, and is a file system encapsulation loss setting determined based on actual measurements of highway wide area network communication.
[0153] For example, during the dynamic unloading of a task, the heartbeat response time is 5000 milliseconds. Based on actual measurements of highway WAN communication, the preset timeout threshold is set to 2000 milliseconds. Calculate the network connection status parameters: Assuming a preset interruption threshold of 1, and the current network connection status parameter indicating a network interruption, the system activates the local policy database of the edge nodes. Based on actual measurements of highway WAN communication, the basic instruction value is set to 10, and the autonomous adjustment coefficient is set to 0.8. The system obtains a service flow classification label of 7. Calculate the offline autonomous control instruction: The system controls the operation of local critical services according to offline autonomous control commands, collecting 100 megabytes of business transaction data. Based on actual measurements using a highway wide area network, the compression ratio is set to 0.4, and the storage overhead factor is set to 1.05. Local incremental cache data is calculated as follows: .like Figure 3 As shown, this illustrates the popularity distribution relationship between the latency requirement feature level and the interaction intensity feature level in business flow characteristics, which together map to generate business flow classification labels. It can be seen that as latency requirements and interaction intensity increase, the generated business flow classification label level increases accordingly, thereby ensuring that services with high latency sensitivity and high concurrency interaction receive higher resource scheduling priority.
[0154] Optionally, the method further includes:
[0155] The network connection status parameters are continuously monitored, and when network recovery is detected, the local incremental cache data is obtained.
[0156] The local incremental cache data is timestamped and prioritized to generate a breakpoint resume data stream;
[0157] The interrupted data stream is smoothly and adaptively adjusted based on the current network bandwidth parameters and synchronized to the upper-level central node.
[0158] Specifically, the system continuously monitors network connection status parameters. When these parameters fall below a preset recovery threshold, the network link is reconnected. At this point, the system reads local incremental cache data from local storage. The local incremental cache data undergoes timestamp verification and priority sorting. Timestamp verification checks the absolute time order of data block generation to prevent logical errors. Priority sorting determines the sending order of data blocks based on service flow classification tags. After verification and sorting, a breakpoint-resume data stream is generated. This breakpoint-resume data stream represents a queue of data to be sent, reorganized according to strict timing and importance. The calculation formula is:
[0159] ;
[0160] in, The quantized priority volume of the breakpoint resumed data stream is measured in megabytes. This represents locally incremental cached data, measured in megabytes. This represents the timestamp verification coefficient, which is dimensionless and is set based on the time-series validity weight determined by actual measurements of highway network recovery. Represents the business flow classification label, dimensionless. Obtains the current network bandwidth parameters after recovery. The current network bandwidth parameters represent the actual available data transmission rate of the physical link at this time. Performs smooth adaptive adjustment on the interrupted data stream based on the current network bandwidth parameters. Smooth adaptive adjustment refers to dynamically changing the data transmission frequency according to the real-time bandwidth to avoid instantaneous congestion. After adjustment, the system synchronizes the data to the upper-level central node. This adjustment process is achieved by calculating the synchronization adjustment rate:
[0161] ;
[0162] in, This represents the synchronization rate, measured in seconds. The quantized priority volume of the breakpoint resumed data stream is measured in megabytes. This represents the current network bandwidth parameter, measured in megabytes per second. Represents the smoothing control factor, which is dimensionless and is set based on the flow shaping buffer ratio determined by actual measurements during highway network recovery.
[0163] For example, continuously monitoring network connection status parameters, when network recovery is detected, the system acquires 42 megabytes of local incremental cache data. The system acquires a service flow classification label of 7. Based on actual measurements of highway network outage recovery, the timestamp verification coefficient is set to 1.2. The system performs timestamp verification and priority sorting on the local incremental cache data, and calculates the quantized priority volume of the resumed data stream: Megabytes per second. The current network bandwidth parameter is 50 megabytes per second. Based on actual measurements of highway network outage recovery, the smoothing control factor is set to 1.5. The data stream for breakpoint resumption is smoothly and adaptively adjusted according to the current network bandwidth parameter, and the synchronization adjustment rate is calculated: Seconds. Data was synchronized to the upstream central node at this rate, successfully completing the breakpoint resume operation.
[0164] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides a hyper-converged system collaborative computing management system for multi-level edge nodes of highways, the system comprising:
[0165] The resource abstraction module is used to obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing.
[0166] The health analysis module is used to obtain the operation index parameters of the multi-level edge nodes, perform health quantification analysis on the operation index parameters, and generate a real-time availability index for the nodes.
[0167] The business classification module is used to acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels.
[0168] The topology construction module is used to obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, the module combines the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels to construct a multi-level edge topology structure diagram.
[0169] The collaborative optimization module is used to input the multi-level edge topology graph into the collaborative unloading model for multi-objective optimization processing and generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm.
[0170] The adaptive correction module is used to migrate the target computing task to the target edge node for processing according to the task dynamic unloading strategy, generate task execution feedback parameters, and adaptively correct the collaborative unloading model based on the task execution feedback parameters.
[0171] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.
[0172] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.
Claims
1. A collaborative computing management method for hyperconverged systems of multi-level edge nodes in highways, characterized in that, The method includes: Obtain heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing; Obtain the operational indicator parameters of the multi-level edge nodes, perform health metric analysis on the operational indicator parameters, and generate a real-time availability index for the nodes. Acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels; Obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, construct a multi-level edge topology diagram by combining the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels. The multi-level edge topology graph is input into the collaborative unloading model for multi-objective optimization processing to generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm. According to the dynamic task unloading strategy, the target computing task is migrated to the target edge node for processing, task execution feedback parameters are generated, and the collaborative unloading model is adaptively corrected based on the task execution feedback parameters.
2. The collaborative computing management method for hyper-converged systems of multi-level edge nodes in highways according to claim 1, characterized in that, The acquisition of heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, through resource abstraction processing, includes: Obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of devices with different architectures in the multi-level edge nodes of the highway to obtain the underlying heterogeneous resource parameters; The underlying heterogeneous resource parameters are containerized and encapsulated into lightweight virtual machines to generate standardized virtual resource parameters; The standardized virtual resource parameters are aggregated across nodes to generate edge computing resource pool parameters.
3. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 1, characterized in that, Obtaining the operational metrics parameters of the multi-level edge nodes and performing health metric analysis on the operational metrics parameters includes: Obtain the hardware operating status data and environmental monitoring data of the multi-level edge nodes to obtain operating index parameters; The operating index parameters are subjected to latent fault prediction feature extraction to generate a hardware degradation trend feature vector. The real-time availability index of a node is generated by combining the hardware degradation trend feature vector with a preset utility evaluation function for multi-dimensional weighted calculation.
4. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 1, characterized in that, The process of acquiring traffic flow data generated at the edge of the highway and performing multi-level classification processing on the traffic flow data includes: Obtain raw traffic flow data generated at the edge of the highway, perform feature parsing on the raw traffic flow data, and generate traffic flow feature parameters; Based on the latency requirement and interaction intensity features in the business flow characteristic parameters, the original business flow data is clustered to generate business flow clusters. Priority mapping is performed on the business flow clusters to generate business flow classification labels.
5. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 1, characterized in that, Obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed a preset load threshold, construct a multi-level edge topology diagram by combining the edge computing resource pool parameters, the node real-time availability index, and the service flow classification label, including: Obtain the real-time computing request data of the multi-level edge nodes and calculate the local load parameters; When the local load parameter exceeds the preset load threshold, the edge computing power resource pool parameter and the node real-time availability index are extracted as the node feature matrix. By combining the node feature matrix, the service flow classification labels, and the network connection relationships between nodes, a multi-level edge topology diagram is constructed.
6. The collaborative computing management method for hyper-converged systems of multi-level edge nodes in highways according to claim 1, characterized in that, The process of inputting the multi-level edge topology diagram into the collaborative unloading model for multi-objective optimization includes: The multi-level edge topology graph is input into a graph neural network for graph node representation learning to generate a global network state feature vector. The global network state feature vector is input into a distributed game theory optimization algorithm, and the algorithm iteratively solves the problem with the goal of minimizing processing latency and maximizing resource balance to generate a task allocation matrix. The task allocation matrix is parsed to generate a dynamic task unloading strategy.
7. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 1, characterized in that, According to the dynamic task unloading strategy, the target computing task is migrated to the target edge node for processing, task execution feedback parameters are generated, and the collaborative unloading model is adaptively corrected based on the task execution feedback parameters, including: According to the dynamic task unloading strategy, the target computing task is scheduled to the target edge node for processing; Obtain the actual latency and energy consumption parameters of the target edge node after executing the target computing task, and generate task execution feedback parameters; A gradient update function is constructed based on the task execution feedback parameters to adaptively adjust the network weight parameters of the graph neural network and the distributed game theory optimization algorithm.
8. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 1, characterized in that, The method further includes: During the dynamic unloading and execution of the task, the network connection status parameters between the multi-level edge nodes and the upper-level central node are obtained; When the network connection status parameter indicates a network interruption, the local policy database of the edge node is activated to generate offline autonomous control instructions; According to the offline autonomous control instructions, the operation of local critical business is controlled, and the business flow data is compressed and stored to generate local incremental cache data.
9. The hyper-converged system collaborative computing management method for multi-level edge nodes of highways according to claim 8, characterized in that, The method further includes: The network connection status parameters are continuously monitored, and when network recovery is detected, the local incremental cache data is obtained. The local incremental cache data is timestamped and prioritized to generate a breakpoint resume data stream; The interrupted data stream is smoothly and adaptively adjusted based on the current network bandwidth parameters and synchronized to the upper-level central node.
10. A hyperconverged system collaborative computing management system for multi-level edge nodes of highways, characterized in that, The system is used for the collaborative computing management method of hyper-converged systems for multi-level edge nodes of highways as described in any one of claims 1-9, and the system includes: The resource abstraction module is used to obtain the heterogeneous computing resource parameters, storage resource parameters, and network bandwidth parameters of multi-level edge nodes of highways, and generate edge computing power resource pool parameters through resource abstraction processing. The health analysis module is used to obtain the operation index parameters of the multi-level edge nodes, perform health quantification analysis on the operation index parameters, and generate a real-time availability index for the nodes. The business classification module is used to acquire business flow data generated at the edge of the highway, perform multi-level classification processing on the business flow data, and generate business flow classification labels. The topology construction module is used to obtain the local load parameters of the multi-level edge nodes. When the local load parameters exceed the preset load threshold, the module combines the edge computing power resource pool parameters, the real-time availability index of the nodes, and the service flow classification labels to construct a multi-level edge topology structure diagram. The collaborative optimization module is used to input the multi-level edge topology graph into the collaborative unloading model for multi-objective optimization processing and generate a dynamic task unloading strategy. The collaborative unloading model includes a graph neural network and a distributed game theory optimization algorithm. The adaptive correction module is used to migrate the target computing task to the target edge node for processing according to the task dynamic unloading strategy, generate task execution feedback parameters, and adaptively correct the collaborative unloading model based on the task execution feedback parameters.
