Network data evaluation methods, electronic devices and storage media based on edge computing

Abstract

This invention provides a network data evaluation method, electronic device, and storage medium based on edge computing. The method includes: deploying a distributed monitoring agent module in an edge computing node cluster to construct a data transmission path topology map for each node; using the path topology map and dynamically labeling physical link attributes; synchronously collecting a raw database of performance indicators for transmission links between nodes; filtering abnormal links by applying a sliding window analysis health factor to the raw performance indicator database; establishing a first healthy target node sequence list for the currently suspicious links; analyzing the comprehensive performance score of nodes experiencing sudden congestion on the currently abnormal links based on the raw performance indicator database, and then establishing a second healthy target node sequence list based on the comprehensive performance score; updating the links in the first healthy target node sequence list based on the second healthy target node sequence list, and triggering a resource reallocation instruction to the edge scheduler.

Description

Technical Field

[0001] This invention relates to the field of edge computing, and more particularly to a network data evaluation method, electronic device, and storage medium based on edge computing. Background Technology

[0002] With the development of network technology, especially the rise of edge computing, more and more applications and services require more efficient and low-latency computing and data transmission between terminal devices and data centers. Edge computing, by distributing computing tasks from centralized data centers to network edge nodes closer to users and terminal devices, not only reduces data transmission latency but also alleviates the burden on central nodes, improving the overall performance and response speed of the system.

[0003] Because edge nodes are typically located in different physical locations and connected to various network devices and other computing nodes, their performance and reliability are affected by a variety of factors, such as network bandwidth, latency, packet loss rate, and link quality. Furthermore, existing edge computing network management systems often have limitations in performance monitoring, especially in multi-node cluster environments, making it difficult to accurately and in real-time assess and optimize the transmission link performance between individual nodes.

[0004] Traditional methods typically acquire network data through centralized performance monitoring. However, this approach encounters issues such as data loss, latency, and accuracy problems in large-scale distributed networks, making real-time performance optimization difficult. Meanwhile, dynamically updating the link topology and performance metric database better reflects changes in network status and allows for comprehensive evaluation and optimization of nodes experiencing sudden congestion, ensuring efficient network operation.

[0005] While existing technologies have improved performance evaluation methods for edge computing networks to some extent, several challenges remain. For example, how to achieve efficient dynamic link reconfiguration in large-scale networks, and how to quickly and accurately identify and adjust abnormal nodes during sudden congestion. These issues remain a significant challenge for achieving efficient, stable, and reliable edge computing networks. Summary of the Invention

[0006] The purpose of this invention is to provide a network data evaluation method, electronic device, and storage medium based on edge computing, which solves the aforementioned technical problems pointed out in the prior art.

[0007] This invention provides a network data evaluation method based on edge computing, comprising the following steps:

[0008] Deploy a distributed monitoring agent module in the edge computing node cluster, with one distributed monitoring agent module set up for each edge computing node; construct a data transmission path topology map for each node; and dynamically label the physical link attributes using the path topology map.

[0009] The system synchronously collects a raw database of performance indicators for transmission links between nodes; it then performs a sliding window analysis of the health factor on the raw database to identify suspicious links; it filters abnormal links from the set of nodes corresponding to the current suspicious links using the health factor; it creates a first healthy target node sequence list for the current suspicious links; and it analyzes the comprehensive performance score of nodes during sudden congestion periods on the current abnormal links based on the raw database of performance indicators, and then establishes a second healthy target node sequence list based on the comprehensive performance score.

[0010] The link is updated based on the second health target node sequence list, and a resource reallocation instruction is triggered to the edge scheduler.

[0011] Secondly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the above-mentioned method steps when executing the program stored in the memory.

[0012] Thirdly, the present invention provides a storage medium comprising a stored program, wherein the program, when running, controls the device where the storage medium is located to execute the aforementioned network data evaluation method based on edge computing.

[0013] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0014] Analysis of the network data evaluation method, electronic device, and storage medium based on edge computing provided by this invention reveals that, in practical applications, the data of each node is first monitored by setting a fixed sampling period to capture packet header feature data, reducing the processing of the entire data packet; data is classified and stored through two independent cache units to ensure effective management of latency and bandwidth-sensitive data, and is quickly transmitted to the original database of performance indicators through a direct memory access (DMA) channel; the health of the link is evaluated by using a sliding window method, combined with weighted coefficients such as latency, packet loss rate, and jitter, forming an instantaneous health score for each link node; the decline rate is calculated by comparing the scores of three consecutive sliding windows, and suspicious links that may have problems are screened out to ensure dynamic real-time evaluation of the network; for suspicious links, the influence factors of each link node are analyzed, and the stability of the node is evaluated by the fluctuation amplitude of the maximum and minimum influence factors, so as to promptly identify unstable nodes; by setting a fluctuation threshold, the system can determine which nodes are unstable nodes, form a node-level health matrix, and intervene in advance to prevent potential failures;

[0015] Furthermore, calculating the health matrix of all nodes and identifying unstable nodes can determine critical paths, thereby efficiently locating performance bottlenecks in the network. The mapping relationship between the health score of each critical path and the node impact factor further helps to understand the contribution of each node to the overall path health. By calculating the mean and standard deviation of the health score of each critical path and introducing a dynamic threshold y, the scheme can flexibly respond to changes in network health status. The introduction of the dynamic threshold makes the judgment of path health more adaptive, avoiding overly strict or lenient evaluation criteria, thus enabling timely detection and screening of abnormal links and early warning of potential problems in the network. For abnormal links, by extracting the starting node, intermediate nodes, and terminal nodes from the path topology graph to form a node set V, the impact factor, minimum health score, and remaining available resource rate of each node are further analyzed and weighted to obtain the repair priority of each node. The repair priority of a node is based on its impact on the overall network health. Prioritizing the repair of nodes with the greatest impact on health helps to achieve reasonable resource allocation, maximize network stability, and generate a first-health target node order list.

[0016] Furthermore, by monitoring key metrics such as transmission latency, packet loss rate, bandwidth utilization, and computing resource utilization, network administrators can monitor the real-time operational status of each node. These metrics reflect the health status of the nodes, helping to identify service interruptions and prevent network congestion or outages. Calculating reliability decay factors based on the number of service interruptions allows for further analysis of each node's stability and recovery capabilities. Calculating the reliability coefficient of each node reveals the speed and effectiveness of its recovery during failures, providing data support for network optimization. Calculating the differences in disaster recovery saturation between different nodes identifies resource-constrained and overloaded nodes, preventing some nodes from becoming network bottlenecks due to excessive load and causing overall performance degradation. A comprehensive performance score for each node is derived by comprehensively evaluating its reliability, adaptability, and disaster recovery capabilities. These scores provide a basis for optimizing the network structure, generating a second list of target healthy nodes to ensure the continuous availability of network services. Attached Figure Description

[0017] Figure 1 Here is a flowchart of the main process of a network data evaluation method based on edge computing in Example 1;

[0018] Figure 2 This is a flowchart illustrating the generation of a sequence list based on a node-level health matrix in an edge computing-based network data evaluation method according to Embodiment 1.

[0019] Figure 3 This is a schematic diagram of the node i influence factor in a network data evaluation method based on edge computing, as described in Example 1.

[0020] Figure 4 This is a main flowchart of the first healthy target node sequence of a network data evaluation method based on edge computing in Embodiment 1;

[0021] Figure 5 This is a schematic diagram of an abnormal link in a network data evaluation method based on edge computing, as described in Embodiment 1.

[0022] Figure 6 This is a main flowchart of the second healthy target node sequence list of a network data evaluation method based on edge computing in Embodiment 1;

[0023] Figure 7 This is a schematic diagram of the structure of a storage medium for applying the above-mentioned network data evaluation method based on edge computing.

[0024] Labels: Memory 1130; Communication interface 1120; Processor 1110; Computer storage medium 1140; Communication bus 1150. Detailed Implementation

[0025] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0026] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0027] Example 1

[0028] like Figure 1 As shown, this embodiment of the invention provides a network data evaluation method based on edge computing, including the following steps:

[0029] S1: Deploy a distributed monitoring agent module in the edge computing node cluster, and set up a distributed monitoring agent module for each edge computing node; construct a data transmission path topology map for each node; and dynamically label the physical link attributes using the path topology map.

[0030] It should be noted that a monitoring agent module is deployed on each edge computing node to enable real-time monitoring and data collection of the performance of each node and between nodes; the above distributed deployment can ensure that the performance of each node is monitored individually, avoiding the problems of data loss or delay.

[0031] The above steps, by constructing a cross-node data transmission path topology map, clearly present the connection relationships between various nodes in the network and update the physical link attributes in real time (i.e., bandwidth and latency, packet loss rate and jitter, error rate, link capacity, etc.). Bandwidth refers to the rate at which a link can transmit data, usually expressed in units such as bits per second (bps), kilobits per second (kbps), or megabits per second (Mbps). Bandwidth determines the maximum data transmission capacity of a link; latency is the time required for data to be transmitted from the source node to the destination node.

[0032] Packet loss rate indicates the proportion of data packets lost during data transmission. Packet loss can be caused by poor link quality, network congestion, or equipment failure; jitter refers to the instability of data packet arrival times. Significant jitter can degrade the quality of real-time applications such as voice and video calls; error rate refers to the proportion of bits or data packets that are erroneous during transmission. A high error rate leads to data retransmission, impacting link performance; link capacity is the maximum data throughput a link can support within a specific time period, usually closely related to bandwidth. It reflects the link's processing capability.

[0033] Physical medium type refers to the type of medium through which data is transmitted, such as optical fiber, cable, and radio waves. Different physical media have different impacts on link performance; for example, the bandwidth of an optical fiber link is usually higher than that of a copper cable link. Transmission distance refers to the physical length of the link, affecting signal attenuation and latency. Long-distance transmission may lead to signal attenuation, requiring signal amplification or retransmission. Link reliability refers to the stability and availability of the link during long-term operation. A highly reliable link can guarantee fewer interruptions and failures. Redundancy refers to designing additional backup paths or equipment in the link to ensure that the system can automatically switch to the backup link when a link fails, avoiding network interruption. It can help identify bottlenecks, delays, and potential congestion points in data flow.

[0034] S2: Synchronously collect the raw database of performance indicators of transmission links between nodes; obtain suspicious links by performing sliding window analysis on the raw database of performance indicators and health factors; filter abnormal links for the node set corresponding to the current suspicious link by health factors; establish a first healthy target node sequence list for the current suspicious link; analyze the comprehensive performance score of the nodes in the sudden congestion period of the current abnormal link based on the raw database of performance indicators, and then establish a second healthy target node sequence list based on the comprehensive performance score.

[0035] It should be noted that by synchronously collecting transmission performance data between nodes, the integrity and timeliness of the data are ensured. The original index database is a database that stores various performance data, which is usually used to record and manage various performance indicators related to the system, network, application, hardware device, etc. These performance indicators reflect the system's operation over a certain period of time, including but not limited to resource usage, response time, throughput, bandwidth utilization, latency, error rate, etc.

[0036] Real-time correlation analysis of the raw performance index database enables rapid identification of abnormal links in the network, ensuring timely problem location and response; a first-healthy target node sequence list and a second-healthy target node sequence list are compiled; the comprehensive performance scores of these nodes during sudden congestion are analyzed using the raw performance index database to determine their reliability under the current link environment; by analyzing the raw performance index database of nodes, nodes that perform well during sudden congestion can be identified, providing a reliable reference for subsequent link reconstruction.

[0037] The first and second health target node order lists help optimize resource allocation and link adjustment, ensuring network health and load balancing.

[0038] S3: Update the link based on the first healthy target node sequence list based on the second healthy target node sequence list, and trigger a resource reallocation instruction to the edge scheduler.

[0039] It should be noted that step S2 above analyzes the comprehensive performance score of nodes during sudden congestion periods. This scoring process combines multi-dimensional performance data of nodes to help evaluate key characteristics such as the stability of each node. The scoring process is a holistic reflection of the health status of nodes. Based on the comprehensive performance score and the health target node order list, the data transmission links are recombined to ensure that traffic passes through the healthiest and most reliable nodes in order to solve the bottleneck problems that may occur in the links under high load or abnormal conditions.

[0040] After sending a resource reallocation instruction to the edge scheduler, the edge scheduler performs the resource reallocation process, which ensures that the system can dynamically adjust resources, avoid overload, and improve the overall network performance. The above process, through link reorganization and resource scheduling, can optimize data flow in real time to avoid network congestion and improve network efficiency. At the same time, resource reallocation helps to achieve load balancing, ensuring that the computing resources of edge computing nodes are not over-concentrated, thereby improving the fault tolerance and reliability of the system.

[0041] Specifically, such as Figure 2 As shown, in step S2, a raw database of performance indicators for transmission links between nodes is synchronously collected; by performing a sliding window analysis of the health factor on the raw database of performance indicators, suspicious links are obtained; the set of nodes corresponding to the current suspicious links is filtered for abnormal links using the health factor; a first healthy target node sequence list is generated for the current suspicious links; based on the raw database of performance indicators, the comprehensive performance score of nodes during the sudden congestion period is analyzed for the current abnormal links, and then a second healthy target node sequence list is established based on the comprehensive performance score. The specific operation steps are as follows:

[0042] Step S21 mainly describes setting a sampling period for each node in the path topology to capture packet header feature data; analyzing the packet header feature data to obtain latency-sensitive and bandwidth-sensitive indicator data; and transmitting the latency-sensitive and bandwidth-sensitive indicator data to the performance indicator raw database through a direct memory access channel. The specific steps are as follows:

[0043] S21: Set a sampling period for each link in the path topology diagram, monitor the physical link attributes of data transmission at each node according to the sampling period, and capture packet header feature data;

[0044] Two independent cache units are pre-established within each node;

[0045] Based on the two independent cache units, the packet header feature data captured by each node is analyzed to obtain latency-sensitive and bandwidth-sensitive indicator data.

[0046] The latency-sensitive index data and the bandwidth-sensitive index data are stored in the two independent cache units respectively to obtain the latency-sensitive unit and the bandwidth-sensitive unit;

[0047] An acceleration module is built for each node, and the acceleration module is used to configure direct memory access (DMA) channels for the latency-sensitive units and bandwidth-sensitive units of each node.

[0048] The storage data in the latency-sensitive unit and bandwidth-sensitive unit is transferred to the performance index raw database using the direct memory access (DMA) channel.

[0049] It should be noted that a fixed sampling period (e.g., every 1 millisecond, 5 milliseconds, or 10 milliseconds) is set in the system configuration beforehand; a filtering strategy is used to capture only the header data required for data transmission at each edge node, avoiding processing the entire data packet and thus reducing system overhead; the header feature data of the data transmission packet includes: source IP and destination IP, transport layer port information, protocol type (TCP / UDP, etc.), timestamp or reception delay, and other customized link identification information; these header feature data are organized and classified according to a predefined format;

[0050] The classification is based on two pre-established independent cache units on each node, categorized into latency-sensitive and bandwidth-sensitive data. This categorized data may cause load issues or other adverse effects on the nodes. Therefore, this categorized data needs to be transferred via Direct Memory Access (DMA) channels according to the storage areas (i.e., latency-sensitive units and bandwidth-sensitive units). The transfer is accelerated by a module and quickly transferred to the historical database for future use. Latency-sensitive data includes information such as packet transmission latency and queuing latency; bandwidth-sensitive data includes metrics directly related to data volume, such as packet size, traffic rate, and packet drop rate.

[0051] Steps S22-S23 mainly describe the process of applying a sliding window to each link, standardizing the performance indicators of nodes within each sliding window using the original database, calculating the instantaneous health score of each node as a health factor, and using the monitoring agent module to calculate the rate of decrease in scores between three consecutive sliding windows to filter out suspicious links. The specific steps are as follows:

[0052] S22: Define a delay weight coefficient, a packet loss rate weight coefficient, and a jitter weight coefficient for each link;

[0053] A sliding window is created for each link, and each sliding window is set according to a pre-defined sampling period (i.e., a sliding window is created on each link, for example, the window size covers 5 sampling periods of data, and after the window ends, it slides for one sampling period, and the sampling period in step S21 is used as the size of the sliding window; that is, if the sampling period of each node is 1, then each sliding window may cover 5 nodes).

[0054] The performance index raw database of each node in the sliding window is standardized, and the standardized performance index raw database is analyzed to calculate the scores of latency weight coefficient, packet loss rate weight coefficient and jitter weight coefficient.

[0055] The instantaneous health score of each node in each link is calculated based on the score of the node within each sliding window, and is used as a health factor.

[0056] It should be noted that the weighting coefficients are defined as weighting coefficients for latency, packet loss rate, jitter, etc., for each link; according to the set sampling period, the sliding window method is used to continuously monitor the link performance, with each window covering multiple sampling periods; the sliding window method ensures dynamic and real-time evaluation of network performance; through the sliding window, the health status of the link can be tracked in real time, and potential problems can be detected in a timely manner.

[0057] After standardization, a health score for each link is obtained by calculating scores for various performance indicators; the health status of the link is comprehensively evaluated through different indicators (latency, packet loss, jitter); by using multiple indicators, we avoid relying on only a single performance data and ensure a more comprehensive health assessment.

[0058] S23: Use the monitoring agent module to compare the scores of three consecutive sliding windows and calculate the rate of decrease in the score between each sliding window;

[0059] A preset continuous decline rate threshold q is set; it is determined whether the decline rate of the three consecutive sliding windows is greater than the continuous decline rate threshold q (that is, whether the decline rate of the three consecutive sliding windows exceeds the continuous decline rate threshold q).

[0060] If so, the link will be classified as a suspicious link;

[0061] It should be noted that the scores of three consecutive sliding windows are compared and the decline rate is calculated. If the decline rate of the scores in the three consecutive windows is greater than the set threshold, the link is judged to be suspicious. The continuous decline in scores indicates that the health of the link is deteriorating rapidly, which may indicate a fault or performance problem. Early detection of possible link problems can avoid a significant drop in network performance. The above step S23 can quickly locate the link that may have a problem by analyzing the trend of score changes.

[0062] Steps S24-25 mainly describe the process of randomly selecting a node i and its adjacent nodes in the path topology graph for each suspicious link; calculating the influence factor of node i using the instantaneous health score of each suspicious link and the actual bandwidth of each suspicious link; calculating the fluctuation amplitude using the maximum and minimum values ​​of the influence factor of node i; and filtering unstable nodes by fluctuation amplitude to form a node-level health matrix. The specific steps are as follows:

[0063] S24: For each suspicious link, randomly select a node i, obtain the neighboring nodes of node i in the path topology graph, and form a neighbor node set N(i).

[0064] For each suspicious link, extract the instantaneous health score of the corresponding node and simultaneously obtain the actual bandwidth of each suspicious link;

[0065] The influence factor of node i is calculated using the instantaneous health score of each node on each suspicious link and the corresponding actual bandwidth.

[0066] It should be noted that when calculating the influence factor of node i, the link performance indicators between i and all its directly connected neighbor nodes will be taken into account.

[0067] By traversing the N(i) process, the health score and bandwidth of each link (from i to neighboring node j) can be accumulated, thereby reflecting the influence of node i on the overall network health status. Then, using N(i), we can more intuitively understand the position and role of node i in the network, that is, how it is connected to other nodes, and whether it is a critical node or a high-risk node.

[0068] For each suspicious link, the impact factor of node i is calculated and analyzed in conjunction with the performance indicators of neighboring nodes. Then, the link performance between node i and its neighboring nodes is analyzed to assess its impact on the overall network health. Step S24 above, by calculating the impact factor, understands the impact of each node on the overall network health, helping to identify key nodes (i.e., the actual bandwidth collected is also affected by some latency, jitter, etc., which affect the network health of node i, hence the impact factor of node i is obtained). The impact factor is used to reflect the network bandwidth, jitter, and latency, and the impact of these factors on the load caused by the subsequent generation of the sequence list in the network. Figure 3 (as shown)

[0069] At the same time, it not only analyzes the node itself, but also considers its connection status with other nodes to comprehensively assess its impact; through comprehensive analysis of neighboring nodes, it can more accurately determine the node's contribution to network health; and help identify key nodes in the network to avoid the overall network being affected by a single node problem.

[0070] S25: Monitor each node i according to a preset recording time, and record the maximum and minimum influence factors of node i within the preset recording time.

[0071] The percentage difference is calculated using the maximum and minimum impact factors, and this value is used as the fluctuation range.

[0072] A preset fluctuation threshold r is set; it is then determined whether the fluctuation amplitude exceeds the fluctuation threshold r.

[0073] If so, then node i is determined to be an unstable node, and a node-level health matrix is ​​formed;

[0074] It should be noted that the monitoring of maximum and minimum impact factors records the maximum and minimum impact factors of a node within a set time period and calculates the fluctuation range. If the fluctuation range exceeds the threshold, the node is determined to be an unstable node. Assessing the stability of a node helps to identify nodes with fluctuations and prevent potential failures. By detecting unstable nodes, timely intervention can be carried out to ensure the long-term stable operation of the network.

[0075] By monitoring the fluctuation amplitude, unstable factors can be detected in a timely manner, preventing them from developing into serious problems; and by providing early warnings of unstable nodes, network design can be optimized and the occurrence of failures can be reduced.

[0076] S26: Utilize the path topology graph to find critical paths for all unstable nodes and calculate path health scores; construct a path-level health matrix using the path health scores and the influence factors of the corresponding node i; count unstable nodes for each critical path in the path-level health matrix to find abnormal links; construct a first healthy target node sequence list based on the abnormal links; analyze the sudden congestion performance data of the nodes on the abnormal links, and calculate a comprehensive performance score for each node using the sudden congestion performance data; sort the nodes according to their comprehensive performance scores to obtain a second healthy target node sequence list.

[0077] It should be noted that the above steps involve identifying all unstable nodes on the critical path and calculating their path health scores; focusing on unstable nodes and critical paths ensures that the parts most affecting network health are addressed promptly; the above steps optimize the health status of paths through critical path health assessment to avoid network paralysis; the above steps analyze nodes on abnormal links, calculate comprehensive performance scores, and rank them; the steps optimize resource allocation and improve overall network performance through comprehensive performance scores; and using scores and rankings, identify which links most need optimization to avoid resource waste.

[0078] Specifically, such as Figure 4As shown, in step S26, the critical paths of all unstable nodes are found using the path topology graph, and the path health score is calculated; a path-level health matrix is ​​constructed using the path health score and the influence factor of the corresponding node i; unstable nodes are counted for each critical path in the path-level health matrix to find abnormal links; a first healthy target node sequence list is constructed based on the abnormal links; the sudden congestion performance data of the nodes on the abnormal links is analyzed, and a comprehensive performance score is calculated for each node using the sudden congestion performance data; the nodes are sorted according to their comprehensive performance scores to obtain a second healthy target node sequence list. The specific operation steps are as follows:

[0079] S261: Calculate the shortest path for data transmission among all unstable nodes in the node-level health matrix of the path topology graph, and use it as the critical path;

[0080] For each critical path, obtain the instantaneous health score of the node corresponding to the suspicious link;

[0081] The instantaneous health scores of nodes on suspicious links corresponding to the critical path are used to count unstable nodes, and the total instantaneous health scores of the unstable nodes are used as the path health score of the critical path.

[0082] It should be noted that the health matrix of all nodes in the path topology graph is calculated to find the shortest data transmission path for each unstable node and identify it as the critical path; then, the instantaneous health score of each suspicious link in the critical path is obtained, and the health of unstable nodes is statistically analyzed based on this.

[0083] Calculating the shortest path helps identify the critical paths that most directly impact network health. By combining the health score of each node with the health score of the critical path, it is possible to accurately assess the performance bottlenecks of the entire network and help prioritize the resolution of critical nodes and links that may affect the health of the entire system. By focusing on addressing issues on the critical path, the overall health of the network can be improved efficiently, and resource allocation and management can be optimized.

[0084] S262: Establish a mapping relationship between the path health score of each critical path and the influence factor of each node i corresponding to the critical path to form a path-level health matrix;

[0085] It should be noted that by mapping the path health score of each critical path to the influence factor of each node on that path, a path-level health matrix is ​​formed. This matrix helps analyze the influence of each node on the path and its contribution to the overall path health. Mapping the path health score of each critical path to the influence factor of each node i on the critical path allows for a clear understanding of each node's influence on the critical path. Furthermore, by associating the influence of nodes with the health of the critical path, it is possible to identify which nodes occupy a more important position on the critical path, enabling more targeted network optimization.

[0086] Steps S263-S265 mainly describe how to set a dynamic threshold y by calculating the mean and standard deviation of the path health scores of all critical paths; how to judge the path health score of each critical path using the dynamic threshold y to filter abnormal links; how to calculate the repair priority of each node based on the remaining available resources of each node in the abnormal links; and how to determine the repair priority of each node based on the repair priority of each node to generate a first healthy target node order list. The specific steps are as follows:

[0087] S263: Calculate the mean and standard deviation of the path health score for all critical paths in the path-level health matrix;

[0088] The dynamic threshold y is calculated by introducing an adaptive adjustment factor based on the mean and standard deviation.

[0089] Determine whether the path health score of each critical path is less than the dynamic threshold y;

[0090] If so, count the number of unstable nodes in the node-level health matrix on the critical path. If the number of unstable nodes is greater than 1, then the critical path is considered an abnormal link.

[0091] It should be noted that the mean and standard deviation of the path health score of all critical paths are calculated, and then an adaptive adjustment factor is introduced based on the mean and standard deviation to calculate the dynamic threshold y; then, it is determined whether the path health score of each critical path is lower than the dynamic threshold y, and whether it is abnormal is determined accordingly.

[0092] The mean and standard deviation of the path health scores mentioned above can serve as a "yardstick" for network health. The introduction of dynamic thresholds makes the judgment criteria more adaptable, better able to respond to changes in network health status, and helps identify paths with lower-than-expected health, thus detecting potential problems early. Dynamic thresholds can adjust the warning criteria based on real-time data fluctuations, ensuring that health assessments adapt to changes in network conditions and promptly identify abnormal links, such as... Figure 5 As shown, avoid overly strict or lenient evaluation criteria;

[0093] S264: Extract the starting node, intermediate nodes (i.e., nodes passed through), and terminal node from the path topology graph for each abnormal link to form a node set V;

[0094] For each node in each node set V, obtain the influence factor (i.e., the acquisition steps are as described in S24 and will not be repeated) and the instantaneous health score of the minimum node;

[0095] For each node in the node set V, calculate the remaining available resource rate (i.e., the capacity that can still be used, which is a prior art and will not be elaborated here).

[0096] The repair priority of a node is obtained by weighting the impact factor of each node in each node set V with the minimum instantaneous health score of the node and the remaining available resource rate.

[0097] It should be noted that for each abnormal link, the starting node, intermediate node and terminal node in the path topology graph are extracted to form a node set V; then, the impact factor, minimum instantaneous health score of the node and remaining available resource rate of each node are obtained, and the repair priority of each node is calculated by combining these factors.

[0098] By obtaining the node set V and calculating the repair priority, we can accurately assess which nodes have the greatest impact on network health in abnormal links, and thus prioritize the repair of those critical nodes that affect the overall health of the network. Furthermore, the repair priority helps to prioritize nodes based on their actual impact and resource availability, avoiding resource waste and ensuring that the most important nodes are repaired first, thereby guaranteeing network stability.

[0099] S265: Sort each node in descending order according to its repair priority, determine the repair priority of each node according to the descending order, and generate a first healthy target node order list;

[0100] It should be noted that the repair priority of each node is determined by sorting them in descending order according to their repair priority, and a first healthy target node order list is generated. The above steps S264-S265, through sorting and priority setting, can ensure that the repair work is carried out in the order of the most critical nodes. In this way, the health of the network can be maximized, while avoiding wasting too many resources on secondary nodes.

[0101] S266: Analyze the sudden congestion performance data for each abnormal link using the original performance index database; analyze the number of service interruptions for each abnormal link using the sudden congestion performance data to obtain a reliability coefficient; construct an index matrix using the reliability coefficient; calculate the feature vector set of the index matrix to obtain a comprehensive performance score; construct a second healthy target node sequence list using the comprehensive performance score of each node.

[0102] It should be noted that the raw performance index database is used to analyze the sudden congestion performance data of each abnormal link, analyze the number of service interruptions and calculate the reliability coefficient; then, the index matrix is ​​constructed using the coefficient, and the feature vector set of the matrix is ​​calculated to finally obtain the comprehensive performance score of each node and generate the second healthy target node order list.

[0103] Further research revealed that sudden congestion performance data is directly related to node stability. By analyzing this data, potential stability issues can be identified. The above execution steps, through comprehensive performance scoring and reliability coefficient analysis, can more accurately evaluate the performance and stability of each node in the face of sudden congestion. The generated healthy target node order list helps optimize subsequent repair strategies and resource allocation.

[0104] Example 2

[0105] like Figure 1 As shown, this embodiment of the invention provides a network data evaluation method based on edge computing, including the following steps:

[0106] S1: Deploy a distributed monitoring agent module in the edge computing node cluster, and set up a distributed monitoring agent module for each edge computing node; construct a data transmission path topology map for each node; and dynamically label the physical link attributes using the path topology map.

[0107] S2: Synchronously collect the raw database of performance indicators of transmission links between nodes; obtain suspicious links by performing sliding window analysis on the raw database of performance indicators and health factors; filter abnormal links for the node set corresponding to the current suspicious link by health factors; establish a first healthy target node sequence list for the current suspicious link; analyze the comprehensive performance score of the nodes in the sudden congestion period of the current abnormal link based on the raw database of performance indicators, and then establish a second healthy target node sequence list based on the comprehensive performance score.

[0108] S3: Update the link based on the first healthy target node sequence list based on the second healthy target node sequence list, and trigger a resource reallocation instruction to the edge scheduler.

[0109] The processing steps in this second embodiment are exactly the same as those in the first embodiment. The difference is that, based on the first embodiment, it also implements a detailed process scheme for calculating the comprehensive performance score.

[0110] Research has shown that by comprehensively evaluating the multi-dimensional performance of each node (such as transmission latency, data packet loss rate, bandwidth utilization, and computing resource utilization), a node health scoring system can be constructed. Furthermore, analyzing the number of service outages and their recovery speed for each node, as well as assessing the node's stability and resilience during sudden congestion or other anomalies, can reveal the node's stability and resilience. Calculating the performance differences between adjacent nodes can determine the degree of interdependence between nodes, reducing load conflicts and impacts between them. Simultaneously, analyzing the node's disaster recovery capabilities can identify which nodes have greater fault tolerance in the face of sudden congestion and are better able to cope with high loads and system pressure.

[0111] At the same time, expanding the second health target node order list helps improve the network's adaptability, enabling it to respond better to emergencies and adjust flexibly in the face of load changes; specifically, such as Figure 6 As shown, in step S266, the sudden congestion performance data of each abnormal link is analyzed using the original performance index database; the number of service interruptions of each abnormal link is analyzed using the sudden congestion performance data to obtain a reliability coefficient; an index matrix is ​​constructed using the reliability coefficient; a feature vector set is calculated on the index matrix to obtain a comprehensive performance score; and a second healthy target node sequence list is constructed using the comprehensive performance score of each node. The specific operation steps are as follows:

[0112] In steps S2661-S2667, the main steps are to analyze the sudden congestion performance data of the first k nodes of each abnormal link using the original performance index database; calculate the number of service interruptions ζ of the sudden congestion performance data of the first k nodes according to the preset timestamp; and calculate the reliability coefficient b based on the square root of the number of service interruptions ζ of the first k nodes and the preset node congestion vulnerability repair level.

[0113] Based on the sudden congestion performance data, the maximum disaster recovery resource Q_max of each of the top k nodes is analyzed to obtain the disaster recovery saturation λ of each node; the difference in disaster recovery saturation λ between each adjacent node of the top k nodes is calculated to obtain the performance impact factor between nodes. The fault recovery rate ω of each node is analyzed by the maximum disaster recovery resource Q_max to obtain the tolerance parameter τ of each node; the reliability coefficient b is updated by the disaster recovery saturation λ and then combined with the tolerance parameter τ to construct the index matrix M;

[0114] Principal component analysis was used to perform dimensionality reduction analysis on the index matrix M to obtain the eigenvector set Φ; the inter-node performance influence factors were then used. The feature vector set Φ is smoothed and then combined with the tolerance parameter τ to calculate the comprehensive performance score of each node;

[0115] Based on the comprehensive performance score of each node, sort them in descending order to establish a sequential list of second healthy target nodes. The specific steps are as follows:

[0116] S2661: Analyze the burst congestion performance data of the first k nodes of each abnormal link using the latency-sensitive index data and bandwidth-sensitive index data from the original performance index database;

[0117] Collect the burst congestion performance data of the first k nodes of each abnormal link according to the preset timestamp;

[0118] The burst congestion performance data includes transmission delay (Δt), packet loss rate (δ), bandwidth utilization (β), and computing resource utilization (η).

[0119] It should be noted that the latency-sensitive and bandwidth-sensitive metrics in the original performance metric database represent data such as transmission data latency and packet drop rate (as explained in step S21 and will not be repeated here). These metrics can be used to analyze the transmission latency (Δt), packet loss rate (δ), bandwidth utilization (β), and computing resource utilization (η) of the first k nodes of each abnormal link. Research has shown that when network nodes experience sudden data congestion, the first k nodes of the abnormal link are often the first to experience congestion or abnormality. Moreover, the congestion of these k nodes is caused by transmission data latency (Δt), packet loss rate (δ), bandwidth utilization (β), and computing resource utilization (η, i.e., insufficient node capacity or computing abnormalities causing congestion).

[0120] By centrally monitoring and analyzing the key performance indicators of the top k nodes, network bottlenecks or sudden congestion can be identified earlier, allowing for timely intervention and repair, and avoiding greater service interruptions.

[0121] S2662: Calculate the number of occurrences of sudden congestion performance data for the first k nodes of each abnormal link based on the most recent n time points in the preset timestamps, and use this as the number of service interruptions ζ;

[0122] The square root of the number of service interruptions ζ for the first k nodes is calculated, and the reciprocal ξ is obtained as the reliability attenuation factor ξ.

[0123] Preset node congestion vulnerability remediation levels: v1, v2, v3;

[0124] The reliability coefficient b is calculated using the node congestion vulnerability repair level v and the reliability decay factor ξ.

[0125] It should be noted that when a sudden congestion performance data occurs among the first k nodes within a certain time period (i.e., a time point n in the preset total timestamps), it indicates that the nodes are congested and may stop transmitting, which is a service interruption. At this time, the number of service interruptions within that time point is recorded as ζ (i.e., the reliability decay factor ξ is used to represent the degree to which the reliability of a system, device, or component gradually decreases during use; over time, the performance of a device or system may degrade due to wear, aging, environmental factors, or other reasons, and the degree of this degradation can be quantified by the decay factor).

[0126] Based on the collected performance data, the number of service interruptions ζ of the top k nodes of each abnormal link in the most recent n time points is calculated; then, the square root of these service interruption counts ζ is processed to obtain the reciprocal ξ, which is used as the reliability decay factor; then, the reliability coefficient b of the node is calculated by combining the node congestion vulnerability repair level v and the decay factor ξ.

[0127] The reliability coefficient b is used to measure the frequency of service outages of a node over a period of time. Frequent outages may indicate that the node is less reliable, while less frequent outages indicate that it is more stable. By using the reciprocal to correct the square root of the number of service outages, the speed and extent of the node's recovery when it encounters an outage can be effectively reflected. Combining the repair level and the attenuation factor, a more accurate reliability assessment is obtained, which further helps to judge the stability of the node.

[0128] The combination of service outage frequency and recovery capability mentioned above can quantify the stability and reliability of nodes, which is very important for network optimization and disaster recovery design.

[0129] S2663: Calculate the maximum disaster recovery resource Q_max for each of the first k nodes based on the sudden congestion performance data (that is, how much capacity can still bear the load after the first k nodes cause congestion based on the sudden congestion performance data. Subtract the sudden congestion performance data of each node to obtain the maximum disaster recovery resource. The "quantity" reflects the highest operating load that the system can withstand in the event of a disaster).

[0130] The monitoring agent module is used to collect the actual processing volume Q_real of sudden node congestion in the history of the first k nodes.

[0131] The disaster recovery saturation λ of each of the first k nodes is calculated using the maximum disaster recovery resource Q_max and the actual processing volume Q_real.

[0132] Calculate the difference in disaster recovery saturation λ between each of the first k nodes and their adjacent nodes to obtain the inter-node performance impact factor. ;

[0133] It should be noted that the maximum disaster recovery resource quantity Q_max is calculated for the first k nodes, and the actual processing quantity Q_real of each node is obtained by the monitoring agent module, and the disaster recovery saturation λ of each node is further calculated; by calculating the difference between the disaster recovery saturation λ of adjacent nodes, the performance impact factor between nodes is obtained.

[0134] The disaster recovery resource Q_max represents the maximum load that a node can withstand under sudden congestion, and is an indicator of the node's resilience. The disaster recovery saturation λ is calculated for each node, which can clearly show the current load status of the node, whether it is close to saturation, and whether there is an overload risk. The performance impact factor represents the performance difference between nodes, reflecting the influence and dependency between different nodes in the network.

[0135] By monitoring the disaster recovery resources and actual processing capacity of the nodes mentioned above, it is possible to identify which nodes will encounter bottlenecks under sudden loads, which helps to make reasonable allocation of system resources in advance.

[0136] S2664: The reliability coefficient κ is corrected using the disaster recovery saturation λ to obtain the reliability coefficient b' of each of the first k nodes;

[0137] The actual processing volume Q_real is used to analyze the sudden congestion performance data to determine the node congestion recovery speed, and the fault recovery rate ω of each of the top k nodes is obtained.

[0138] The tolerance parameter τ of each of the first k nodes is obtained by using the fault recovery rate ω and the maximum disaster recovery resource amount Q_max.

[0139] The reliability coefficient b' and tolerance parameter τ in the first k nodes are standardized to construct the index matrix M;

[0140] It should be noted that the reliability coefficient b is corrected based on the disaster recovery saturation λ of the first k nodes to obtain the updated reliability coefficient b'; at the same time, the fault recovery rate ω of each node is calculated in combination with the actual processing volume Q_real, and the tolerance parameter τ is further calculated.

[0141] The updated reliability coefficient b', through the correction of disaster recovery saturation, ensures that the reliability assessment can more realistically reflect the performance of the node under different loads; the fault recovery rate ω measures the node's recovery capability and helps to evaluate the time and efficiency of the node to recover to normal state after a failure; the tolerance parameter τ measures the node's adaptability and recovery capability under abnormal conditions and is an important evaluation indicator of the node's overall performance.

[0142] The updated reliability coefficient and tolerance parameters can comprehensively reflect the stability and recovery capability of nodes in the event of emergencies;

[0143] S2665: Use principal component analysis to sort the tolerance parameters τ in the index matrix M in descending order (that is, sort the tolerance parameters τ as the contribution rate of the nodes, because the tolerance parameters τ include the nodes' performance in congestion and recovery, so they are analyzed as the contribution rate of principal component analysis), and extract the principal components of the first h nodes with the highest tolerance parameters τ as the feature vector set Φ.

[0144] Using inter-node performance impact factors Smooth the eigenvectors of node a in the eigenvector set Φ, and then smooth the smoothed node h (i.e., the inter-node performance influence factor). It is the influence factor between adjacent nodes ac, so it can smooth the feature vectors of the first h nodes with tolerance parameter τ, that is, update node a); iterates through each node in the feature vector set Φ to perform the inter-node performance influence factor. The smoothing process yields an updated set of feature vectors Φ.

[0145] It should be noted that Principal Component Analysis (PCA) is used to sort the tolerance parameters τ of the nodes, and the principal components of the first h nodes with the highest tolerance parameters τ are extracted as the set of eigenvectors Φ. By smoothing these eigenvectors, the impact of performance differences between neighboring nodes is reduced.

[0146] The tolerance parameter τ in the indicator matrix M uses PCA to reduce the dimensionality of the node features and extract the most representative features, which helps to reduce computational complexity and focus the analysis on key nodes; the feature vector of node a in the feature vector set Φ is smoothed to ensure that the node performance will not be affected by the anomalies of adjacent nodes.

[0147] The tolerance parameter τ in the indicator matrix M reduces the data dimensionality and smooths the interaction between nodes through principal component analysis (PCA), making the evaluation results more accurate and operable.

[0148] S2666: Utilize the updated feature vector set Φ and the inter-node performance impact factor The tolerance parameter τ is used to calculate the overall performance score of the node, and the calculation formula is as follows:

[0149] ;

[0150] In the formula, This means that the values ​​of each dimension in the updated feature vector are weighted and aggregated according to certain weights to obtain a preliminary node performance score (referred to as the baseline score).

[0151] Represented as nodes In the The values ​​on the features (i.e., the feature vectors are extracted by principal component analysis (PCA), and obtained after data dimensionlessization, dynamic weighting, neighborhood smoothing, and tolerance parameter adjustment; they reflect the comprehensive effect of a node in a certain dimension (which may correspond to latency, packet loss rate, bandwidth utilization, computing resource utilization, etc.), and due to data standardization, the dimensions can be directly compared).

[0152] Represented as weighting factor (i.e. Indicates assigning the first The weights of the feature vectors are used to reflect the importance of the feature in the overall score; they are often calculated using the following normalization method: ;in This is the corrected reliability coefficient, reflecting the relationship with the first... Dimensionally relevant node reliability or stability information; the arctan function can be used to handle large or abnormal [data / information]. The values ​​are smoothed to limit their impact on the weights; the denominator ensures that all... The sum of these features is 1, which makes the contribution of each feature in the final aggregation proportional (the sum of these features is 1).

[0153] This is represented as the adjustment factor (i.e., used to perform a secondary correction on the weighted aggregation result so that the comprehensive performance score simultaneously reflects the node's own "tolerance" and its performance coupling with neighboring nodes).

[0154] Represented as nodes The tolerance parameter is typically calculated from factors such as the proportion of redundant node resources (e.g., spare resources) and the fault recovery rate (i.e., After Sigmoid function form After mapping, its value can be normalized to a smooth and finite range (usually between 0 and 2); the tolerance parameter reflects the node's adaptability and recovery ability when encountering anomalies or sudden congestion; a higher tolerance parameter This will cause the score to approach 2, thus improving the overall score; conversely, it will decrease the score.

[0155] This is expressed as the maximum performance impact factor (i.e. Represents a node The maximum performance impact factor in its neighborhood, i.e., the node's performance metric. Performance consistency with its most relevant (or most affected) neighbors; this is typically calculated by constructing a spatiotemporal correlation model, such as calculating the disaster recovery saturation and performance metric differences of neighboring nodes. ;in This represents the load or disaster recovery saturation of a node, and its maximum value is taken as the threshold. ;pass Perform mapping, when A larger value indicates a significant performance difference between the node and its key neighbors; this value tends to decrease, thus having a suppressive effect on the overall score. A lower value indicates a lower performance difference. A consistent performance between a node and its neighbors is beneficial for improving the score; conversely, a poorly consistent performance may indicate that the node is susceptible to anomalies in its neighborhood, resulting in a deduction of points.

[0156] In the above formula, the overall meaning of the adjustment factor is: This expression will give the node's final comprehensive score. The composition is achieved through weighted feature aggregation. Partially, reflecting the nodes The intrinsic fusion evaluation across various feature dimensions captures both the original and preprocessed performance data. A secondary correction is applied to the intrinsic evaluation after adjusting factors, ensuring that nodes are evaluated not only on their own data but also considering their robustness against interference and external coupling effects. The final comprehensive score is the product of these two evaluations. It achieves a dual integration of internal performance and external environmental influences, thereby providing a more comprehensive and dynamic evaluation of node performance;

[0157] It should be noted that a comprehensive performance score for each node is calculated by weighting and aggregating feature vectors and adjusting factors to obtain a baseline score; this score comprehensively considers multiple factors such as node reliability, adaptability, and external influences.

[0158] The weighted aggregation of each dimension described above yields a comprehensive performance score for each node, making the evaluation results more comprehensive. This comprehensive performance score helps to further rank the nodes, assisting network administrators in identifying which nodes require the most optimization and attention.

[0159] Meanwhile, the comprehensive performance score not only considers the performance of a single dimension, but also combines multiple factors such as the adaptability and resilience of nodes, providing a more comprehensive basis for network optimization.

[0160] S2667: Sort each node in descending order according to the comprehensive performance score, and establish a second healthy target node sequence list by sorting.

[0161] It should be noted that, based on the comprehensive performance score of each node, they are sorted in descending order to form a second healthy target node sequence list; by sorting, priority can be given to the worst performing nodes, and optimization or repair measures can be implemented based on their comprehensive scores; by sorting the comprehensive scores of nodes, it can be ensured that network administrators can reasonably allocate resources according to priorities and improve the overall network performance.

[0162] Example 3

[0163] This invention provides an electronic device, including a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1150. The processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1150. The memory 1130 is used to store computer programs. When the processor 1110 executes the program stored in the memory, it implements the steps of the aforementioned network data evaluation method based on edge computing.

[0164] Example 4

[0165] On the other hand, this fourth embodiment, based on the network data evaluation method based on edge computing provided in the first embodiment of the invention, also provides a computer storage medium 1140 (hereinafter referred to as the storage medium). Figure 7 The diagram shown is a schematic of a computer storage medium structure framework provided in Embodiment 3 of the present invention, which includes:

[0166] Memory 1130 is used to store computer programs;

[0167] The communication interface 1120 is used to connect the memory 1130 to the processor 1110;

[0168] Processor 1110 is configured to execute a computer program to implement an embodiment of a network data evaluation method based on edge computing, as disclosed in any combination of the above embodiments.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A network data evaluation method based on edge computing, characterized in that, The following steps are included: Deploy a distributed monitoring agent module in the edge computing node cluster, and set up a distributed monitoring agent module for each edge computing node; construct a data transmission path topology map for each node; Based on the path topology map, the physical link attributes of each link are dynamically labeled; The system synchronously collects a raw database of performance indicators for transmission links between nodes; it then performs a sliding window analysis of the health factor on the raw database of performance indicators to identify suspicious links; it further filters the set of nodes corresponding to the current suspicious links using the health factor to identify abnormal links; and finally, it generates a first healthy target node sequence list for the current suspicious links. Based on the original database of the performance indicators, the comprehensive performance score of the nodes during the sudden congestion period of the abnormal link is analyzed, and a second healthy target node sequence list is established based on the comprehensive performance score. The link is updated based on the first health target node sequence list based on the second health target node sequence list, and a resource reallocation instruction is triggered to the edge scheduler; The raw database of performance metrics for transmission links between nodes is collected synchronously. The specific steps are as follows: A sampling period is set for each node of each link in the path topology to capture packet header feature data; the packet header feature data is analyzed to obtain latency-sensitive and bandwidth-sensitive indicator data, and the latency-sensitive and bandwidth-sensitive indicator data are transmitted to the performance indicator raw database through a direct memory access channel; By performing a sliding window analysis on the original database of performance indicators to identify health factors and screen suspicious links, the specific steps are as follows: A sliding window is applied to each link, and the performance indicators of the nodes within each sliding window are standardized from the original database to calculate the instantaneous health score of each node, which serves as a health factor. The monitoring agent module is used to calculate the rate of decrease in the score between three consecutive sliding windows to identify suspicious links. For the current suspicious link, the set of nodes is filtered for abnormal links using health factors; for the current suspicious link, a first healthy target node sequence list is generated; Based on the original database of performance indicators, the comprehensive performance score of the nodes during the sudden congestion period is analyzed on the abnormal link. Based on the comprehensive performance score, a second healthy target node sequence list is established. The specific operation steps are as follows: For each suspicious link, a node i and its adjacent nodes in the path topology graph are randomly selected; the influence factor of node i is calculated using the instantaneous health score of each node on each suspicious link and the actual bandwidth of each suspicious link. The fluctuation amplitude is calculated by using the maximum and minimum values ​​of the influence factor of node i; unstable nodes are screened by the fluctuation amplitude to form a node-level health matrix. The path health score is calculated by finding the critical paths of all unstable nodes using the aforementioned path topology graph. Construct a path-level health matrix using the path health score and the influence factor of the corresponding node i; count unstable nodes for each critical path in the path-level health matrix to find abnormal links; construct a first health target node sequence list based on the abnormal links. Analyze the sudden congestion performance data of the nodes in the abnormal link, and calculate a comprehensive performance score for each node based on the sudden congestion performance data; The nodes are sorted according to their overall performance scores to obtain the second healthy target node order list.

2. The network data evaluation method based on edge computing according to claim 1, characterized in that, The critical paths of all unstable nodes are found using the aforementioned path topology graph, and path health scores are calculated. A path-level health matrix is ​​constructed using these path health scores and the influence factors of the corresponding node i. The specific steps are as follows: The path topology is used to calculate the shortest path for data transmission among all unstable nodes in the node-level health matrix, which is then used as the critical path; for each critical path, the instantaneous health score of the node corresponding to the suspicious link is obtained. The instantaneous health scores of nodes on suspicious links corresponding to the critical path are used to count unstable nodes, and the total instantaneous health scores of the unstable nodes are used as the path health score of the critical path. A path health matrix is ​​formed by mapping the path health score of each critical path to the influence factor of each node i corresponding to the critical path.

3. The network data evaluation method based on edge computing according to claim 2, characterized in that, For each critical path in the path-level health matrix, count unstable nodes to identify abnormal links; construct a first health target node sequence list based on the abnormal links. The specific steps are as follows: A dynamic threshold y is set by calculating the mean and standard deviation of the path health scores of all critical paths; the path health score of each critical path is judged by the dynamic threshold y to filter abnormal links; the repair priority of each node is calculated by the remaining available resource rate of each node in the abnormal links; the repair priority of each node is determined by the repair priority of each node, and a first healthy target node order list is generated.

4. The network data evaluation method based on edge computing according to claim 3, characterized in that, Analyze the sudden congestion performance data of the nodes on the abnormal links, and calculate a comprehensive performance score for each node based on the sudden congestion performance data; sort the nodes according to their comprehensive performance scores to obtain a second healthy target node order list. The specific operation steps are as follows: The system analyzes the burst congestion performance data for each abnormal link using the original performance index database; it then analyzes the number of service interruptions for each abnormal link using the burst congestion performance data to obtain a reliability coefficient; and finally, it constructs an index matrix using the reliability coefficient. The feature vector set of the index matrix is ​​calculated to obtain a comprehensive performance score; a second healthy target node sequence list is constructed based on the comprehensive performance score of each node.

5. The network data evaluation method based on edge computing according to claim 4, characterized in that, The system analyzes the burst congestion performance data for each abnormal link using the original performance index database; it then analyzes the number of service interruptions for each abnormal link using the burst congestion performance data to obtain a reliability coefficient; and finally, it constructs an index matrix using the reliability coefficient. The feature vector set is calculated on the index matrix to obtain a comprehensive performance score; a second healthy target node sequence list is constructed based on the comprehensive performance score of each node. The specific operation steps are as follows: The performance index database is used to analyze the sudden congestion performance data of the first k nodes of each abnormal link; the number of service interruptions ζ of the sudden congestion performance data of the first k nodes is calculated according to the preset timestamp; the reliability coefficient b is obtained by calculating the square root of the number of service interruptions ζ of the first k nodes and the preset node congestion vulnerability repair level. Based on the sudden congestion performance data, the maximum disaster recovery resource Q_max of each of the top k nodes is analyzed to obtain the disaster recovery saturation λ of each node; the difference in disaster recovery saturation λ between each adjacent node of the top k nodes is calculated to obtain the performance impact factor between nodes. ; The fault recovery rate ω of each node is analyzed by the maximum disaster recovery resource Q_max, and the tolerance parameter τ of each node is obtained; the reliability coefficient b is updated by the disaster recovery saturation λ and then combined with the tolerance parameter τ to construct the index matrix M; Principal component analysis was used to perform dimensionality reduction analysis on the index matrix M to obtain the set of eigenvectors Φ. Using the inter-node performance impact factor The feature vector set Φ is smoothed and then combined with the tolerance parameter τ to calculate the comprehensive performance score of each node; The nodes are sorted in descending order based on their overall performance score to create a sequential list of second-healthy target nodes.

6. An electronic device, characterized in that, The system includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements the steps of the method described in any one of claims 1-5.

7. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device where the storage medium is located to perform the network data evaluation method based on edge computing as described in any one of claims 1-5.

Images