Overlay network path scoring and adaptive routing methods, systems, devices, and media

Abstract

The application provides an overlay network path scoring and adaptive routing method, system, device and medium. The application collects link state and service semantic features of candidate paths on the end side, inputs a lightweight model for forward reasoning, and outputs risk score and stability score of each path. The reasoning process does not interfere with normal transmission of the data plane. In combination with preset rules and dynamic thresholds, the candidate paths are classified, screened and admitted according to two-dimensional scores to obtain an optimized path set. Based on the set, an adaptive routing mechanism executes point-to-point preferential forwarding, relay fallback or multi-path racing strategies, and dynamically adjusts parameters such as the number of concurrent paths and the racing window. At the same time, actual transmission result data is continuously collected, the deviation between the score and the real performance is compared, the score threshold, path priority and routing parameters are iteratively corrected based on the misjudgment statistics, and the closed-loop optimization of the routing strategy is realized, thereby improving the stability, efficiency and dynamic adaptation capability of network transmission.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, system, device, and medium for overlay network path scoring and adaptive routing selection. Background Technology

[0002] In overlay networks, VPNs, P2P / relay hybrid networks, and edge relay networks, candidate path selection often relies on fixed priorities, threshold judgments, or single-instantaneous link observation results. Existing solutions have several drawbacks: they depend solely on fixed thresholds for RTT (Round-Trip Time), packet loss rate, or number of failures, making it difficult to adapt to dynamic changes in different tenants, service priorities, and network environments; instantaneous link metrics cannot predict subsequent link stability; relay path switching is a passive response, triggered only after physical packet loss or disconnection of directly connected links, resulting in lengthy fault recovery times; furthermore, the lack of anti-jitter mechanisms for critical fluctuations easily leads to routing oscillations and service interruptions; path ranking lacks a unified scoring mechanism; and the end-side response to local network changes is slow, hindering timely adaptive path adjustments. Therefore, an effective overlay network path scoring and adaptive routing selection method is urgently needed to address these issues. Summary of the Invention

[0003] In view of the above problems, the present invention is proposed to provide a method, system, device and medium for overlay network path scoring and adaptive routing to overcome or at least partially solve the above problems.

[0004] To achieve the above and other related objectives, the present invention provides a method for coverage network path scoring and adaptive routing selection, the method comprising: The path status features of the candidate path set in the coverage network are collected and extracted by the end side. The path status features include link status features and / or service semantic features. The link status features include at least one of round-trip delay fluctuation, number of consecutive failures, remaining cooldown time, and candidate relay availability. The service semantic features include at least one of budget status or service priority. The path state features are input into a lightweight model deployed on the edge for forward inference to generate path score results for each candidate path; wherein, the path score results include path risk score and / or path stability score, and the generation process does not block hot paths in the original data plane. Based on preset mapping rules and dynamic thresholds, the candidate path set is hierarchically filtered and admitted using the path scoring results to obtain an adjusted candidate path set. Adaptive routing is performed based on the adjusted candidate path set. During the routing process, if the path risk score is higher than the first threshold or the path stability score is lower than the second threshold, a relay fallback mechanism is triggered, and the number of concurrent paths and the racing window in the routing control parameters are dynamically adjusted based on the path score results. The system collects the actual transmission results, backoff times, and recovery time of the actual transmission path, calculates the deviation between the path score and the actual performance to generate misjudgment statistics, and corrects the subsequent path score threshold, candidate path priority, and routing control parameters accordingly to achieve iterative optimization of the routing strategy.

[0005] Optionally, the lightweight model is a machine learning model with limited parameter size, deployed on the edge for inference. Its model weights are offline pre-trained weights or weights obtained by incremental updates based on local historical data on the edge. The continuous training and updating process of the model does not depend on centralized training in the cloud. The generation process of the path score result is asynchronously separated between the edge and the data plane forwarding engine. After the score is generated, the corresponding routing decision instruction is asynchronously sent to the scheduler on the data plane through the state cache or asynchronous queue on the edge.

[0006] Optionally, the step of collecting and extracting path state features from the candidate path set in the overlay network at the edge includes: Link status features and service semantic features of each candidate path in the candidate path set in the coverage network are collected and extracted from the end side. The round-trip delay fluctuations and consecutive failures of each candidate path are correlated and weighted to generate a risk tendency coefficient for each candidate path. The risk propensity coefficients of each candidate path are fused and normalized with the budget status or business priority to generate standardized path status features, which are then used as input to the lightweight model.

[0007] Optionally, the step of using the path scoring results to perform hierarchical screening and admission control on the candidate path set based on preset mapping rules and dynamic thresholds to obtain an adjusted candidate path set includes: The path score results of each candidate path are compared with the preset conditions one by one, and candidate paths with scores lower than the preset conditions are excluded from the path selection or racing range in this round; wherein, the preset conditions are used to determine whether a candidate path is qualified to participate in the selection or racing in this round. The scores of the remaining candidate paths are compared with the path score threshold. The priority of candidate paths with scores below the path score threshold is reduced, and the priority of candidate paths with scores above the path score threshold is increased. Monitor the availability of candidate relays for the remaining candidate paths. If the availability of candidate relays decreases, adjust the admission criteria for the corresponding candidate paths and reduce their priority or remove them from the list. If, after the above adjustments, there are multiple candidate paths with similar scores, then a decision will be made based on business priority, budget status, and path type to determine their relative order. Based on the above adjustments, a dynamically adjusted set of candidate paths is formed.

[0008] Optionally, performing adaptive routing based on the adjusted candidate path set includes: During the execution of the racing mode, when the link operation indicators of the racing path are stable and meet the standards, and the path risk score is continuously lower than the preset security threshold for more than a preset time window, a gradual expansion mechanism is triggered; the number of concurrent paths is gradually increased according to the preset step size and the racing window is linearly widened. After the backup path detection frequency returns to the normal level, the current transmission conditions are re-verified. If both the path risk score and the path stability score of the racing path reach the preset fallback threshold, the racing mode will be smoothly exited and the system will switch to point-to-point forwarding; wherein the preset fallback threshold is higher than the first threshold used to trigger the degradation intervention.

[0009] Optionally, during the route selection process, if the path risk score is higher than a first threshold or the path stability score is lower than a second threshold, a relay fallback mechanism is triggered, including: During the routing process, the path score and link degradation trend of each candidate path are monitored in real time. When the path risk score of a candidate path is detected to be higher than the first threshold, the relay fallback mechanism is immediately triggered to terminate the point-to-point direct forwarding through the candidate path and switch to transmission via the relay path; wherein, the first threshold is a preset high-risk score threshold for the path. If the candidate path synchronously meets the path stability score below the second threshold, then in addition to the relay backoff, a further degradation intervention is performed; wherein, the degradation intervention includes at least one of lowering the transmission priority of the candidate path, marking the candidate path as a cooling state, suspending the candidate path from participating in this round of path racing, and increasing the priority of the backup or relay path; the second threshold is a preset path low stability score threshold.

[0010] Optionally, the dynamic adjustment of the number of concurrent paths and the racing window in the routing control parameters based on the path scoring results includes: When a candidate path is detected to have a preset potential risk, a race suppression action is executed to achieve race suppression and transmission concurrency peak reduction. The preset potential risks include increased fluctuations in path round-trip delay, decreased availability of candidate relay paths, tightening of budget status, or path risk score exceeding a preset safety threshold. The race suppression action includes reducing the number of concurrent paths in this round, tightening the time window for race selection, increasing the frequency of backup path detection, and reducing the participation priority of candidate paths identified as high-risk in this round of race.

[0011] In a second aspect, the present invention also provides a coverage network path scoring and adaptive routing system, the system comprising: a processor and a memory; the memory storing a computer program; the processor being configured to execute the computer program to implement the coverage network path scoring and adaptive routing method as described above.

[0012] Thirdly, the present invention provides an electronic device comprising: at least one processor and a memory communicating with the at least one processor, the memory storing a computer program that, when executed by the processor, causes the electronic device to implement the overlay network path scoring and adaptive routing method as described above; the electronic device further comprising at least one communication interface for interacting with the control plane and data plane of the overlay network; wherein the processor receives service semantic features issued by the control plane through the communication interface and obtains path status features of the candidate path set through a local network module.

[0013] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by instructions of a processor, the processor implements the coverage network path scoring and adaptive routing method as described above.

[0014] Fifthly, the present invention provides a computer program product, which includes computer program code. When the computer program code is run on a computer, the computer implements the coverage network path scoring and adaptive routing selection method as described above.

[0015] The above-described one or more technical solutions provided by this invention can have the following advantages or at least achieve the following technical effects: This invention collects link status and service semantic features of candidate paths at the endpoint, inputs them into a lightweight model for forward inference, and outputs risk and stability scores for each path. This inference process does not interfere with normal data plane transmission. Combining preset rules and dynamic thresholds, candidate paths are hierarchically screened and admitted based on the dual-dimensional scores to obtain an optimized path set. Based on this set, an adaptive routing mechanism executes strategies such as point-to-point priority forwarding, relay backoff, or multi-path racing, and dynamically adjusts parameters such as the number of concurrent paths and the racing window. Simultaneously, actual transmission result data is continuously collected, and the deviation between the scores and actual performance is compared. Based on misjudgment statistics, the scoring threshold, path priority, and routing parameters are iteratively corrected to achieve closed-loop optimization of the routing strategy, thereby improving the stability, efficiency, and dynamic adaptability of network transmission. Attached Figure Description

[0016] Figure 1 The diagram shows a flowchart of a network path scoring and adaptive routing method according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a state transition mechanism based on path scoring-driven relay rollback, cooling, and recovery in one embodiment of the present invention.

[0018] Figure 3 The diagram shown is a schematic representation of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0019] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0020] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0021] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0022] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0023] Unless otherwise stated, the term "multiple" means two or more.

[0024] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0025] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0026] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] Please see Figure 1 An embodiment of the present invention provides a method for coverage network path scoring and adaptive routing selection, the method including the following steps S10~S50: Step S10: The path status features of the candidate path set in the coverage network are collected by the end side; wherein, the path status features include link features and / or service semantic features; the link status features include at least one of round-trip delay fluctuation, number of consecutive failures, remaining cooldown time and candidate relay availability; the service semantic features include at least one of budget status or service priority.

[0028] The term "end-side" refers to the local execution unit in the client device, edge node agent, or user-space network component, which is responsible for performing path scoring and route selection based on the local context.

[0029] The candidate path set refers to the set of all alternative routing paths that can be used for data transmission, pre-selected by the end side based on the current network topology, service requirements, or pre-calculation strategies.

[0030] Path state features are a set of indicator parameters used to characterize the transmission quality and service operation environment of the overlay network, including link state features and / or service semantic features.

[0031] Link state characteristics refer to those used to reflect the operating status of the underlying physical links and relay nodes of the network. They include at least one of the following: round-trip time, packet loss rate, number of consecutive failures, cooling status, and availability of candidate relays.

[0032] Business semantic features refer to those that reflect the needs of upper-layer applications, resource quotas, and user identity attributes. They include at least one of the following: budget status, business priority, tenant identifier, namespace, and service identifier.

[0033] In practical implementation, at least one component of the end-side network node (such as a local route selector, link observation module, control plane metadata, session state cache, or equivalent path state structure) can periodically probe and collect data in real time from the candidate path set in the overlay network, extracting the link state features and service semantic features of each candidate path. The link state features include, but are not limited to, one of the following: round-trip time (RTT), number of consecutive failures (such as the number of consecutive point-to-point failures or consecutive relay failures), cooling status (including remaining cooling time), and candidate relay availability, which are used to reflect the real-time transmission quality and stability of the path. The service semantic features include at least one of the following: budget status and service priority, to associate service requirements with resource constraints. During the collection process, the end-side components need to synchronously perform standardized preprocessing operations, including data cleaning (such as outlier removal), format conversion (unified quantization scale), and feature integration (such as time-series data aggregation), to ensure the standardization of path state information and finally generate standardized path state features.

[0034] Step S20: Input the path state features into a lightweight model deployed on the edge for forward inference to generate path score results for each candidate path; wherein, the path score results include path risk score and / or path stability score, and the generation process does not block hot paths in the original data plane.

[0035] Lightweight models refer to optimized machine learning models deployed on client devices, local agent processes, edge node agents, or user-space network components. They are used to perform inference locally based on link state features and business semantic features and output path score results without uploading feature data to the cloud.

[0036] Path scoring results refer to multi-dimensional evaluation data generated by the lightweight model based on path state characteristics, used to quantitatively evaluate the comprehensive transmission quality and service adaptability of candidate paths; it includes at least one of the following: path risk score, path stability score, path availability score, reachability judgment result, multi-level path priority label, racing window admission suggestion, relay backoff early execution suggestion, concurrent path number adjustment parameter, racing window adjustment parameter, and backup path detection frequency adjustment parameter.

[0037] In a practical implementation, the path state features of each candidate path can be input into a lightweight model deployed on the edge. This model relies on locally pre-trained weights for forward inference calculation. After the calculation is completed, path score results for each candidate path are generated. The path score results include path risk score and / or path stability score. The entire inference process runs independently and does not affect the main flow of original data transmission. That is, the generation process does not block hot paths in the original data plane. Finally, these path score results provide a reliable decision basis for adaptive routing selection on the edge.

[0038] Furthermore, in one embodiment, the lightweight model is a machine learning model with limited parameter size, deployed on the edge for inference. Its model weights are offline pre-trained weights or weights obtained by incremental updates based on local historical data on the edge. The continuous training and updating process of the model does not rely on centralized training in the cloud. The generation process of the path scoring result is asynchronously separated between the edge and the data plane forwarding engine. After the score is generated, the corresponding routing decision instruction is asynchronously sent to the scheduler on the data plane through the state cache or asynchronous queue mechanism on the edge.

[0039] In this embodiment, a lightweight model with a small parameter scale is deployed on the edge, and its initial weights are derived from offline pre-training. When the model needs to continuously adapt, incremental training can be performed using local historical data through lightweight techniques, such as updating only some parameters. This training process is completed entirely locally, without relying on centralized training in the cloud. Network decision-making and data forwarding functions are decoupled: path scoring results are generated by the lightweight model, and the generation process is asynchronously separated from the forwarding engine of the data plane. After the model outputs the path scoring results, it does not directly block the data flow, but stores the corresponding routing decision instructions in the edge-side state cache or asynchronous queue. Subsequently, the data plane scheduler asynchronously obtains the instructions and executes packet forwarding. The lightweight model runs independently as a local inference module, and its output process does not block the hot paths of the original data plane, thus ensuring that the original data transmission is not affected. By inferring latency fluctuation characteristics through the edge-side lightweight model, low-latency path scoring can be generated without blocking the original data plane; latency fluctuation characteristics include, but are not limited to, round-trip latency variance, standard deviation, and other indicators.

[0040] Furthermore, in one embodiment, after the lightweight model outputs the path score result, consistency verification, threshold constraint verification, or correction based on equivalent rules can be performed on the path score result; if the path score result meets the rollback condition, violates the recovery condition, or has abnormal racing parameters, the path score result will be automatically corrected according to preset rules; after the correction is completed, the affected candidate path will be downgraded, and a conservative routing strategy will be enabled for the candidate path.

[0041] Among them, the preset rules refer to the criteria used to automatically verify and correct the path scoring results output by the lightweight model; they mainly cover three types of rules: The first category is routine verification and correction rules: consistency rules to ensure the self-consistency of scoring logic, threshold rules to limit the reasonable range of scoring, and equivalent rules to handle cases where scores are the same or paths are equivalent. The second category is abnormal triggering rules: used to define rollback conditions, which are automatically triggered when performance abnormalities or racing parameters are detected to be seriously deviated from their normality. The third category is conflict resolution rules: which primarily adjudicate conflicts between multiple recovery conditions.

[0042] When the above rules are triggered, the path score results of the affected candidate paths are first automatically corrected according to the corresponding rules; then, the routing policies of these candidate paths are downgraded to the preset conservative policies to ensure the stability and reliability of the overall network.

[0043] Step S30: Based on preset mapping rules and dynamic thresholds, the candidate path set is subjected to hierarchical screening and admission control using the path scoring results to obtain an adjusted candidate path set.

[0044] Among them, the preset mapping rules refer to the pre-set filtering or priority adjustment strategies used to determine candidate paths based on path scoring results.

[0045] As an example, based on business needs or predefined strategies, it can be stipulated that when the path score result is lower than the dynamic threshold, the priority of the path will be automatically reduced; or it can be stipulated that when the path score result is higher than the dynamic threshold, the priority of the path will be automatically increased; ensuring that the system performs path sorting and filtering operations according to the established logic, reducing manual intervention.

[0046] Dynamic threshold refers to the scoring criteria used to determine whether a candidate path is qualified; its value can be adjusted in real time according to the actual situation.

[0047] As an example, when network congestion worsens, this threshold can be dynamically increased to filter out more low-scoring paths; conversely, when network resources are plentiful, the threshold can be dynamically decreased to retain more alternative paths. This dynamic adjustment mechanism allows the system to flexibly adapt to different network conditions, balancing the quantity and quality of paths.

[0048] The adjusted candidate path set refers to the optimized path set that is finally formed after the initial candidate paths have undergone a series of processes such as scoring, verification, correction and priority adjustment, and is used for actual routing decisions.

[0049] In practical implementation, the candidate path set can be initially screened based on the path scoring results using dynamic thresholds, eliminating candidate paths with scores below the current dynamic threshold and retaining only qualified paths that meet the score requirements. Then, qualified paths are finely graded using preset rules, dividing them into different priority levels (such as high priority, medium priority, and alternative priority) according to the scoring range and business strategy, and marking them with corresponding priority labels. Subsequently, admission control is executed based on the grading results, combining real-time resource status (such as bandwidth utilization) and path dynamic characteristics (such as latency fluctuations) to dynamically verify and adopt paths of each priority level, selecting paths that meet the current network conditions and business requirements. Through the above graded screening and admission control based on path scoring results, a final adjusted candidate path set that combines quality, adaptability, and flexibility is formed.

[0050] Step S40: Perform adaptive routing selection based on the adjusted candidate path set; during the routing selection process, if the path risk score is higher than the first threshold or the path stability score is lower than the second threshold, trigger the relay backoff mechanism and dynamically adjust the number of concurrent paths and the racing window in the routing control parameters based on the path score results.

[0051] The first threshold refers to the preset high-risk scoring threshold for the path.

[0052] The second threshold refers to the preset path instability score threshold.

[0053] Routing control parameters refer to the core variables that can be dynamically adjusted during adaptive routing selection and multi-path race transmission. They are used to directly control the execution of path selection strategies, optimize network resource allocation, and enable flexible switching of transmission modes. The core parameters include three categories: number of concurrent paths (dynamically adjusting the number of paths for parallel transmission), race window (dynamically setting the time window length for path performance evaluation), and backup path detection frequency (dynamically setting the health monitoring cycle for backup paths).

[0054] In its implementation, relying on the adjusted candidate path set, the endpoint performs adaptive routing selection based on real-time network status and path score results. Its core is a three-level progressive decision-making process: First, point-to-point direct forwarding is prioritized. When the path risk score of a direct path exceeds the first threshold, or the path stability score falls below the second threshold, a relay fallback mechanism is immediately triggered, switching to a backup relay path to mitigate risk. Second, if the relay path performance also fails to meet transmission requirements, a multi-path racing mechanism is initiated: high-scoring paths are selected from the candidate set, and the number of concurrent paths and the racing window size are dynamically adjusted to race in parallel and select the best-performing transmission path. Throughout the process, path scores are continuously monitored. Once a risk score exceeds the limit or stability is insufficient, a threshold-triggered mechanism is used for proactive adjustments, such as actively reducing traffic allocation to high-risk paths, adjusting racing strategy parameters, and increasing the backup path detection frequency, thereby ensuring timely switching of risky paths and maintaining transmission stability and efficiency. This process achieves dynamic coordination between routing selection, fallback, and racing.

[0055] Step S50: Collect the actual transmission results, backoff times and recovery time of the actual transmission path, calculate the deviation between the path score and the actual performance to generate misjudgment statistics, and correct the subsequent path score threshold, candidate path priority and routing control parameters accordingly to achieve iterative optimization of the routing strategy.

[0056] The actual transmission path refers to the path selected in adaptive routing through a three-level decision-making process (direct connection priority, relay backoff, and multi-path competition). Based on real-time network conditions (latency, bandwidth, packet loss rate) and actual transmission data (sending results, number of backoffs, recovery time, etc.), it dynamically optimizes through scoring deviation correction to ensure that the optimal path is always used in dynamic networks, thus guaranteeing transmission stability and efficiency.

[0057] The actual transmission result, referring to the final status confirmation of data packet transmission, is a key input for path scoring. Success means the data packet is received by the receiving end and validly acknowledged; failure means the data packet is lost or times out, triggering a retransmission or discard mechanism. This result is directly used to calculate the deviation of the path score.

[0058] The number of rollbacks refers to the number of times the routing policy executes a fast switch to a backup relay path when the performance of the current primary path is below a threshold; it is an important statistical measure used to measure path instability and trigger policy adjustments.

[0059] Recovery time refers to the total time from confirming a path failure to rebuilding a stable data flow on the new path; it encompasses fault detection latency, path switching decision and execution time, and the convergence time for the new path traffic to reach the expected performance. This metric directly reflects the system's agility and recovery capability in responding to faults.

[0060] Misjudged statistical data refers to various operational statistical data and status records generated by the transmission path over multiple consecutive operating cycles, which differ from the real-time transmission results collected at the current moment.

[0061] The path scoring threshold is a critical value used in adaptive routing to determine whether a path is usable, and is used to filter low-quality candidate paths. It can be dynamically calculated based on real-time network performance metrics such as latency, bandwidth, or packet loss rate to support continuous optimization of adaptive routing strategies.

[0062] In its implementation, key performance data of the actual transmission path is collected in real time, including actual transmission results, number of backoffs, and recovery time. Historical data is processed through a sliding window or time decay mechanism to focus on recent performance. Then, the deviation between the measured performance and the model score of the path is calculated based on this, and misjudgment data is statistically analyzed. Subsequently, the subsequent path score threshold is dynamically corrected using the misjudgment data to improve the filtering criteria for inefficient paths, simultaneously optimize the priority ranking of candidate paths, and adjust routing control parameters such as the number of concurrent paths, the size of the racing window, and the frequency of backup path detection. By introducing a threshold correction mechanism based on misjudgment statistics and a hysteresis recovery strategy, this closed loop effectively suppresses frequent route switching caused by critical fluctuations in network status, thereby reducing routing oscillations and ensuring the continuous stability and adaptability of data transmission.

[0063] Furthermore, in one embodiment, after the endpoint performs path decision-making based on locally cached path state characteristics, candidate relay availability, or priority path status, if it detects inconsistencies between control plane metadata and the actual data plane status, expired cache timestamps, or the target path being offline, a fast fallback mechanism is triggered, such as switching to an alternative path or enabling relay forwarding mode. The local cache timestamp and candidate path availability status are simultaneously updated, and subsequent path scoring input parameters are adjusted accordingly, such as correcting scoring thresholds or optimizing priority weights. This process is essentially a dynamic adjustment of routing control parameters, using real-time feedback to correct path optimization strategies and transmission mode switching conditions. This strengthens the feedback loop of adaptive routing selection, ensuring that routing strategies can be iteratively optimized, thereby maintaining stability and performance in dynamic networks.

[0064] In this embodiment, the link status and service semantic features of candidate paths are collected at the edge, input into a lightweight model for forward inference, and the risk score and stability score of each path are output. This inference process does not interfere with the normal transmission of data plane. Combining preset rules and dynamic thresholds, candidate paths are graded and screened and admission controlled according to the dual-dimensional scores to obtain an optimized path set. Based on this set, an adaptive routing mechanism executes strategies such as point-to-point priority forwarding, relay backoff, or multi-path racing, and dynamically adjusts parameters such as the number of concurrent paths and racing window. At the same time, actual transmission result data is continuously collected, the deviation between the score and the actual performance is compared, and the score threshold, path priority, and routing parameters are iteratively corrected based on misjudgment statistics to achieve closed-loop optimization of the routing strategy, thereby improving the stability, efficiency, and dynamic adaptability of network transmission.

[0065] Based on the foregoing embodiments, a second embodiment of the network path scoring and adaptive routing method of the present invention is proposed. In this embodiment, step S10 may include the following sub-steps S101~S103: Sub-step S101 involves collecting and extracting the link status features and service semantic features of each candidate path in the candidate path set of the coverage network from the end side.

[0066] In the specific implementation, the components on the end side cooperate to collect the link status characteristics (including round-trip time, packet loss rate, number of consecutive failures, cooling status and candidate relay availability) and business semantic characteristics (covering budget status and business priority) of the candidate paths in real time. These characteristics are then parsed and integrated to form unified path status information, providing real-time and accurate reference for routing decisions.

[0067] Sub-step S102 involves performing a weighted calculation of the round-trip delay fluctuations and consecutive failures of each candidate path to generate a risk tendency coefficient for each candidate path.

[0068] In the specific implementation, based on the link state characteristics of the candidate paths, the risk tendency coefficient of each candidate path is generated by weighting the round-trip time (RTT) fluctuation in the link state characteristics with the number of consecutive failures, so as to quantitatively evaluate the potential transmission risk of the path and provide a reliable basis for routing decisions.

[0069] Sub-step S103 involves fusing and normalizing the risk propensity coefficients of each candidate path with the budget status or business priority to generate standardized path status features, which serve as input to the lightweight model.

[0070] In practical implementation, the risk propensity coefficient of each candidate path can be dynamically integrated with the budget status or business priority, and the Min-Max normalization method is used to unify the dimensions to generate standardized path status features. These features serve as the input of the lightweight model, achieving a quantitative and dynamic balance between link risk and business needs while meeting the model input format requirements.

[0071] In this embodiment, the edge component collects link status features and business semantic features of candidate paths in real time, and dynamically fuses and performs Min-Max normalization on these features to generate standardized path status features, which are then fed into the lightweight model. The lightweight model outputs path scores based on the standardized features, thereby accurately measuring path risk, unifying input data specifications, and improving the reliability of model judgment.

[0072] Based on the foregoing embodiments, a third embodiment of the network path scoring and adaptive routing method of the present invention is proposed. In this embodiment, step S20 may include the following sub-steps S201~S205: Sub-step S201: Compare the path score results of each candidate path with the preset conditions one by one, and exclude candidate paths with scores lower than the preset conditions from the scope of path selection or racing in this round; wherein, the preset conditions are used to determine whether the candidate path is qualified to participate in the selection or racing in this round.

[0073] Among them, the preset conditions refer to the pre-set conditions used to determine whether a candidate path is qualified to participate in the selection or race in this round.

[0074] In practice, by comparing the path score of each candidate path in the candidate path set with the preset conditions, candidate paths that do not meet the score are eliminated, and only paths that meet the preset conditions are selected to enter the subsequent selection or racing stage, thereby optimizing decision-making efficiency and quality.

[0075] Sub-step S202 compares the scores of the remaining candidate paths with the path score threshold, lowers the priority of candidate paths with scores lower than the path score threshold, and raises the priority of candidate paths with scores higher than the path score threshold.

[0076] In practice, the priority of each remaining candidate path can be dynamically adjusted based on the comparison between its score and the path score threshold: paths with scores below the threshold have lower priority, while paths with scores above the threshold have higher priority. This score-based comparison and sorting mechanism can quickly distinguish the relative quality of paths, thus ensuring that high-scoring paths are selected by the scheduler first.

[0077] Sub-step S203: Monitor the availability of candidate relays for the remaining candidate paths. If the availability of candidate relays decreases, adjust the admission criteria for the corresponding candidate paths and reduce their priority or remove them from the list.

[0078] The admission criteria refer to the mandatory requirements that a path must meet before participating in routing selection or competition, covering elements such as minimum performance indicators, maximum risk score limits, minimum stability baselines, and resource reservations. The system only allows paths that fully meet the criteria to enter the subsequent evaluation and allocation stages, thereby ensuring the basic availability and security of the scheduling process from the source.

[0079] In practice, the availability of each relay node in the remaining candidate paths is continuously monitored. Once a decrease in the availability of a relay is detected, the corresponding candidate path is locked, the admission criteria for that path are adjusted, and its priority in this round of competition is reduced. If the relay availability of a path drops too low, the path is removed from the candidate set. Through this dynamic adjustment, unstable paths can be restricted or eliminated in a timely manner, ensuring that the paths ultimately participating in the competition have sufficient reliability.

[0080] In sub-step S204, if multiple candidate paths with similar scores exist after the above adjustments, a decision is made based on business priority, budget status, and path type to determine their relative order.

[0081] Among them, business priority refers to the degree of importance of different business uses; core businesses correspond to higher priority.

[0082] Budget status refers to the cost and expense situation incurred by using this path, including the level of cost and whether the budget is sufficient.

[0083] Path type refers to the category to which the candidate path belongs; such as different transmission lines, relay links, network topology, etc.

[0084] In practical implementation, after the above adjustments, if multiple candidate paths with similar scores are detected, a final comprehensive score is calculated by combining the business priority, current budget status, and path type of these paths. Subsequently, the relative order of these candidate paths is determined based on this comprehensive score. This multi-dimensional evaluation can break the deadlock when scores are similar, ensuring that the final selection not only meets technical standards but also aligns with actual business needs.

[0085] Sub-step S205: Based on the above adjustment results, dynamically form the adjusted candidate path set.

[0086] In the specific implementation, after receiving the processing results of the aforementioned steps, all candidate paths are traversed to check their current status; then, paths that have been removed or do not meet the admission criteria are filtered out, all paths that meet the criteria are retained, and they are reordered according to the latest priority; subsequently, these valid paths are packaged to dynamically generate an adjusted candidate path set; through real-time aggregation and reorganization, it can be ensured that the candidate set always maintains the latest and optimal state, providing accurate basic data for subsequent final path selection.

[0087] In this embodiment, inferior paths that do not meet preset conditions are first screened out; then the priority of the remaining paths is adjusted according to the score; the admission requirements and path levels are updated in real time based on the availability status of relay nodes; for paths with similar scores, they are sorted by comprehensive business priority, budget status and path type; and the list of available paths is dynamically updated overall. This mechanism effectively eliminates high-risk paths, classifies path quality, adapts to real-time network changes, and streamlines the ranking of paths with the same score, thereby improving the overall quality of the final available paths.

[0088] Based on the foregoing embodiments, a fourth embodiment of the network path scoring and adaptive routing method of the present invention is proposed. In this embodiment, step S30 may include the following sub-steps S301~S302: In sub-step S301, during the execution of the racing mode, when the link operation indicators of the racing path are stably up to standard and the path risk score is continuously lower than the preset security threshold for more than a preset time window, a gradual expansion mechanism is triggered; the number of concurrent paths is gradually increased according to the preset step size and the racing window is linearly widened. After the backup path detection frequency returns to the normal level, the current transmission conditions are re-verified.

[0089] Among them, racing paths refer to multiple network paths that simultaneously compete for data transmission in an adaptive routing system.

[0090] Link performance metrics refer to real-time data that directly reflects the performance of network paths and are used to determine whether a path is available; such as RTT, RTT variance, error rate, and bandwidth utilization.

[0091] Preset security thresholds refer to key indicator thresholds that are pre-defined to determine whether a network path is safe and reliable. Examples include the upper limit of path RTT variance, the upper limit of error rate, and the lower limit of risk score.

[0092] A preset time window refers to the shortest time that a pre-defined indicator must continuously meet safety conditions.

[0093] In practical implementation, during the execution of the speed-up mode, when the link operation indicators of the speed-up path are detected to be consistently and stably up to standard, and the path risk score is below the preset security threshold for more than a preset time window, a gradual expansion mechanism will be triggered. Specifically, the number of concurrent paths can be increased in stages with a preset step size, and the range or duration of the speed-up window can be gradually expanded at a preset linear rate. After the concurrent path expansion and speed-up window adjustment are completed, the detection frequency of the backup paths is gradually reduced back to the normal level to reduce resource consumption. After the detection frequency of the backup paths stabilizes, the transmission conditions of all current paths are reassessed (including real-time operation indicators and risk scores, etc.) to ensure that the overall link after expansion still meets security and performance requirements.

[0094] In sub-step S302, if both the path risk score and the path stability score of the racing path reach a preset fallback threshold, the racing mode is smoothly exited and the system is switched to point-to-point forwarding; wherein, the preset fallback threshold is higher than the first threshold used to trigger the degradation intervention.

[0095] The preset rollback threshold refers to a pre-set critical value used to trigger a complete exit from racing mode. This threshold is strictly higher than the first threshold used to trigger a downgrade intervention, so as to ensure that the operation of completely exiting racing mode is only performed when the path risk deteriorates to an unacceptable level over a long period of time.

[0096] In its implementation, when both the path risk score and path stability score of a racing path reach a preset fallback threshold, the system smoothly exits racing mode and switches to point-to-point forwarding. Specifically, it first stops allocating new traffic to the racing path, allowing only established transmission sessions to continue; it then gradually releases backup path resources, closing backup path connections sequentially according to task priority, while maintaining the primary path for continued transmission to ensure uninterrupted service; once the primary path is confirmed to be stable, it shuts down all racing-related processes and resets routing control parameters to their default values ​​for point-to-point forwarding mode, including the number of concurrent paths, racing window, and probe frequency. This anti-oscillation mechanism triggers exit through a dual-indicator joint determination, employing a phased process of stopping traffic allocation, releasing resources, verifying the primary path, and resetting parameters to achieve a seamless exit from racing mode, avoiding service interruptions and traffic surges.

[0097] Furthermore, to improve the reliability of routing, a risk assessment and fallback mechanism can be introduced during the adaptive routing process (i.e., in the specific implementation of step S30), specifically including the following sub-steps S303~S305: Sub-step S303 involves real-time monitoring of the path score and link degradation trend of each candidate path during the routing process.

[0098] In its implementation, the endpoint continuously collects real-time performance metrics for each candidate path, such as Round Trip Time (RTT), packet loss rate, bandwidth utilization, and transmission success rate. Based on these metrics, and considering factors like RTT fluctuations, relay availability, and budget status, a comprehensive score for each path is dynamically calculated and updated in real-time to reflect path health. Subsequently, time-series analysis is used to compare current and historical data to identify link performance degradation trends, such as whether RTT is continuously increasing, packet loss rate is increasing, or bandwidth fluctuations are intensifying, thereby assessing potential risks such as increased RTT variance and sudden changes in packet loss rate. This mechanism emphasizes decision-making based on performance trends rather than instantaneous states to improve the adaptability and reliability of routing.

[0099] Sub-step S304: When the path risk score of a candidate path is detected to be higher than the first threshold, the relay fallback mechanism is immediately triggered to terminate the point-to-point direct forwarding through the candidate path and switch to transmission via the relay path; wherein, the first threshold is a preset path high-risk score threshold.

[0100] In its implementation, the path risk score of the current candidate path is compared with a first threshold. If the score exceeds the first threshold, a fallback process is immediately triggered without waiting for the path to be completely interrupted. The specific fallback process is as follows: Point-to-point direct forwarding through the candidate path is forcibly terminated, new data packets are stopped from being sent to the path, and its transmission channel is closed. Subsequently, a verified and usable relay path is selected from the pre-selection pool, and the data to be transmitted is quickly switched to this path. After the switch is completed, test data packets are sent to confirm the relay path's connectivity, while the operating status of the original point-to-point path is continuously monitored to support subsequent decisions. This relay fallback mechanism, through a closed-loop mechanism of real-time scoring comparison and proactive switching, quickly activates a backup relay path before the path performance deteriorates to the point of unavailability, thereby improving the overall reliability and robustness of transmission.

[0101] In sub-step S305, if the candidate path synchronously meets the path stability score below the second threshold, then on the basis of relay backoff, further degradation intervention is performed; wherein, the degradation intervention includes at least one of lowering the transmission priority of the candidate path, marking the candidate path as a cooling state, suspending the candidate path from participating in this round of path racing, and increasing the priority of the backup or relay path; the second threshold is a preset path low stability score threshold.

[0102] In its implementation, when a candidate path simultaneously meets the criteria of a risk score higher than the first threshold and a stability score lower than the second threshold, a relay fallback mechanism is triggered, along with a degradation intervention. Relay fallback, triggered by the risk score, aims to quickly switch to an alternative relay path to avoid transmission interruption. Degradation intervention, triggered by the stability score, serves as an additional measure to limit path resource usage. This includes lowering the transmission priority of the candidate path, marking it as a "cooling-off" state, suspending its participation in the current path race, and increasing the priority of at least one of the alternative or relay paths. In practice, one or more measures can be dynamically selected. For example, the path can be moved to a low-priority queue to ensure critical traffic uses stable links; it can be marked as "cooling-off" to suspend new task allocation but retain historical records; it can be removed from the current race set to avoid continued use in multi-path concurrent transmission; and the priority of associated alternative paths can be increased to accelerate the switchover. This combined mechanism, triggered by both risk and stability, proactively reduces the impact of problematic paths and optimizes resource allocation on top of ensuring basic connectivity through relay fallback.

[0103] Furthermore, after the relay fallback mechanism is triggered, in order to optimize subsequent transmission, step S30 also performs the following dynamic adjustment operation (sub-step S306): Sub-step S306: When a preset potential risk is detected in a candidate path, a race suppression action is executed to achieve race suppression and transmission concurrency peak reduction. The preset potential risks include increased fluctuations in path round-trip delay, decreased availability of candidate relay paths, tightened budget status, or path risk score exceeding a preset safety threshold. The race suppression action includes reducing the number of concurrent paths in this round, tightening the time window for race selection, increasing the frequency of backup path detection, and reducing the participation priority of candidate paths identified as high-risk in this round of race.

[0104] Among them, the preset potential risks include increased fluctuations in path round-trip delays, decreased availability of candidate relay paths, tightening budget status, or path risk scores exceeding preset safety thresholds.

[0105] In its implementation, when any pre-defined potential risk is detected in a candidate path, such as increased path latency fluctuations, decreased relay availability, budget constraints, or exceeding risk scores, a proactive race-suppression mechanism will be executed. This aims to achieve peak shaving and risk avoidance in concurrent transmission through proactive resource allocation. Specifically, some or all of the following actions will be dynamically executed: reducing the number of paths participating in concurrent transmission in the current round, prioritizing stable and low-risk paths through comprehensive evaluation; tightening the decision-making window for path race selection to expedite decision-making and avoid prolonged exposure in deteriorating environments; increasing the detection and verification frequency of backup paths to ensure rapid switching; and simultaneously reducing the participation priority of candidate paths identified as high-risk in the current race or temporarily suspending their eligibility. This mechanism, through real-time monitoring and threshold judgment of multi-dimensional risk indicators, proactively and hierarchically reduces resource investment in paths before their performance significantly deteriorates or they completely fail, thereby optimizing resource allocation and improving the overall robustness and efficiency of transmission.

[0106] Please see Figure 2 The diagram illustrates the state transition mechanism for relay fallback, cooling, and recovery based on path scoring. This mechanism uses path scoring as the core to drive adaptive closed-loop management of link states, and the specific transition logic is as follows: When a path meets the score and performs stably, it enters a high-score priority state. In this state, the path enjoys the highest priority, and is given priority for point-to-point direct forwarding or inclusion in the race set for transmission.

[0107] When the path risk score rises above the first threshold or the stability score falls below the second threshold, a low-score degradation action is triggered first. Degradation actions include lowering the path priority and suspending participation in the race. If the path performance deteriorates further or the joint triggering conditions are met, the system will force the path to enter a relay fallback state, terminating direct transmission and switching to relay forwarding to maintain basic connectivity.

[0108] After a relay fallback, or when a path is deemed high-risk / unstable, it is proactively placed into a cooling-off protection state. In this state, the path is suspended from participating in new task allocation and only retains historical records to avoid resource waste and frequent network jitter.

[0109] After the cooling-off period ends or after re-detection, the path enters the recovery assessment state; a closed-loop decision is made based on the latest score results: if the score recovers to a safe range, the path returns to the high-score priority state; if the score still fails to meet the standard, the path remains in the relay fallback state or re-enters the degradation and cooling cycle.

[0110] This mechanism, through the linkage of real-time scoring and state machine, achieves full lifecycle management from normal transmission and fault avoidance to automatic recovery, ensuring the robustness of routing selection and the efficiency of resource utilization.

[0111] In this embodiment, under the race-based operation mode, the system dynamically adjusts resources based on real-time path scores. When the long-term performance indicators of a path remain satisfactory and the risk score stays within a safe threshold, race-based expansion is gradually implemented. Specific actions include orderly increasing the number of parallel transmission paths, relaxing the time window for race-based decisions, or increasing the detection frequency of backup paths. After expansion, the stability of the path is verified through active detection, and the carrying capacity of the current transmission environment is comprehensively evaluated to ensure the safe and reliable expansion operation. Conversely, when both the path risk score and stability score are detected to deteriorate simultaneously and reach a preset higher-order backoff threshold, a race-based contraction mechanism is triggered. This process aims for smooth degradation, gradually reducing the number of parallel paths, tightening the race-based strategy, and ultimately orderly switching to a more robust point-to-point transmission mode. This adjustment mechanism achieves real-time matching between transmission strategy and network status. When network conditions are good, race-based expansion effectively improves overall transmission throughput and efficiency; when network conditions deteriorate, timely and smooth contraction ensures the stability of core transmissions, thus achieving a balance between efficiency and robustness.

[0112] Based on the same inventive concept, the fifth embodiment of the present invention also provides a coverage network path scoring and adaptive routing system corresponding to the coverage network path scoring and adaptive routing method of the foregoing embodiments. Since the principle of the system in the fifth embodiment of the present invention is similar to the coverage network path scoring and adaptive routing method of the foregoing embodiments of the present invention, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be described again.

[0113] In addition, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described overlay network path scoring and adaptive routing method.

[0114] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Figure 3 As shown, the electronic device includes at least one processor 401, a memory 402, at least one network interface 403, and a user interface 405. The various components in the electronic device are coupled together via a bus system 404. It is understood that the bus system 404 is used to implement communication between these components. In addition to a data bus, the bus system 404 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 3 The general will label all buses as bus systems.

[0115] The user interface 405 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0116] It is understood that memory 402 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0117] In this embodiment of the invention, the memory 402 is used to store various types of data to support the operation of the electronic device 400. Examples of this data include: any executable program for operation on the electronic device 400, such as the operating system 4021 and application programs 4022; the operating system 4021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 4022 may contain various applications, such as a media player, browser, etc., for implementing various application services. The implementation of the coverage network path scoring and adaptive routing selection method provided in this embodiment of the invention can be included in the application program 4022.

[0118] The methods disclosed in the above embodiments of the present invention can be applied to processor 401, or implemented by processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 401 or by instructions in software form. The processor 401 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 401 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 401 may be a microprocessor or any conventional processor, etc. The steps of the coverage network path scoring and adaptive routing selection method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in a memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0119] In an exemplary embodiment, the electronic device 400 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to perform the aforementioned method.

[0120] In summary, this invention collects link status and service semantic features of candidate paths at the endpoint, inputs them into a lightweight model for forward inference, and outputs risk and stability scores for each path. This inference process does not interfere with normal data plane transmission. Combining preset rules and dynamic thresholds, candidate paths are hierarchically screened and admitted based on the dual-dimensional scores to obtain an optimized path set. Based on this set, an adaptive routing mechanism executes strategies such as point-to-point priority forwarding, relay backoff, or multi-path racing, and dynamically adjusts parameters such as the number of concurrent paths and the racing window. Simultaneously, actual transmission result data is continuously collected, and the deviation between the scores and actual performance is compared. Based on misjudgment statistics, the scoring threshold, path priority, and routing parameters are iteratively corrected to achieve closed-loop optimization of the routing strategy, thereby improving the stability, efficiency, and dynamic adaptability of network transmission.

[0121] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for network path scoring and adaptive routing selection, characterized in that, The method includes: The path status features of the candidate path set in the coverage network are collected and extracted by the end side. The path status features include link status features and / or service semantic features. The link status features include at least one of round-trip delay fluctuation, number of consecutive failures, remaining cooldown time, and candidate relay availability. The service semantic features include at least one of budget status or service priority. The path state features are input into a lightweight model deployed on the edge for forward inference to generate path score results for each candidate path; wherein, the path score results include path risk score and / or path stability score, and the generation process does not block hot paths in the original data plane. Based on preset mapping rules and dynamic thresholds, the candidate path set is hierarchically filtered and admitted using the path scoring results to obtain an adjusted candidate path set. Adaptive routing is performed based on the adjusted candidate path set. During the routing process, if the path risk score is higher than the first threshold or the path stability score is lower than the second threshold, a relay fallback mechanism is triggered, and the number of concurrent paths and the racing window in the routing control parameters are dynamically adjusted based on the path score results. The system collects the actual transmission results, backoff times, and recovery time of the actual transmission path, calculates the deviation between the path score and the actual performance to generate misjudgment statistics, and corrects the subsequent path score threshold, candidate path priority, and routing control parameters accordingly to achieve iterative optimization of the routing strategy.

2. The method according to claim 1, characterized in that, The lightweight model is a machine learning model with limited parameter size, deployed on the edge for inference. Its model weights are either pre-trained offline or incrementally updated based on local historical data on the edge. The continuous training and updating process of this model does not rely on centralized training in the cloud. The generation process of the path score result is asynchronously separated between the edge and the data plane forwarding engine. After the score is generated, the corresponding routing decision instruction is asynchronously sent to the scheduler on the data plane through the state cache or asynchronous queue on the edge.

3. The method according to claim 1, characterized in that, The path state features collected and extracted from the candidate path set in the overlay network by the end side include: Link status features and service semantic features of each candidate path in the candidate path set in the coverage network are collected and extracted from the end side. The round-trip delay fluctuations and consecutive failures of each candidate path are correlated and weighted to generate a risk tendency coefficient for each candidate path. The risk propensity coefficients of each candidate path are fused and normalized with the budget status or business priority to generate standardized path status features, which are then used as input to the lightweight model.

4. The method according to claim 1, characterized in that, The process involves using preset mapping rules and dynamic thresholds to perform hierarchical screening and admission control on the candidate path set based on the path scoring results, resulting in an adjusted candidate path set, including: The path score results of each candidate path are compared with the preset conditions one by one, and candidate paths with scores lower than the preset conditions are excluded from the path selection or racing range in this round; wherein, the preset conditions are used to determine whether a candidate path is qualified to participate in the selection or racing in this round. The scores of the remaining candidate paths are compared with the path score threshold. The priority of candidate paths with scores below the path score threshold is reduced, and the priority of candidate paths with scores above the path score threshold is increased. Monitor the availability of candidate relays for the remaining candidate paths. If the availability of candidate relays decreases, adjust the admission criteria for the corresponding candidate paths and reduce their priority or remove them from the list. If, after the above adjustments, there are multiple candidate paths with similar scores, then a decision will be made based on business priority, budget status, and path type to determine their relative order. Based on the above adjustments, a dynamically adjusted set of candidate paths is formed.

5. The method according to claim 1, characterized in that, The step of performing adaptive routing selection based on the adjusted candidate path set includes: During the execution of the racing mode, when the link operation indicators of the racing path are stable and meet the standards, and the path risk score is continuously lower than the preset security threshold for more than a preset time window, a gradual expansion mechanism is triggered; the number of concurrent paths is gradually increased according to the preset step size and the racing window is linearly widened. After the backup path detection frequency returns to the normal level, the current transmission conditions are re-verified. If both the path risk score and the path stability score of the racing path reach the preset fallback threshold, the racing mode will be smoothly exited and the system will switch to point-to-point forwarding; wherein the preset fallback threshold is higher than the first threshold used to trigger the degradation intervention.

6. The method according to claim 1, characterized in that, During the routing process, if the path risk score is higher than a first threshold or the path stability score is lower than a second threshold, a relay fallback mechanism is triggered, including: During the routing process, the path score and link degradation trend of each candidate path are monitored in real time. When the path risk score of a candidate path is detected to be higher than the first threshold, the relay fallback mechanism is immediately triggered to terminate the point-to-point direct forwarding through the candidate path and switch to transmission via the relay path; wherein, the first threshold is a preset high-risk score threshold for the path. If the candidate path synchronously meets the path stability score below the second threshold, then in addition to the relay backoff, a further degradation intervention is performed; wherein, the degradation intervention includes at least one of lowering the transmission priority of the candidate path, marking the candidate path as a cooling state, suspending the candidate path from participating in this round of path racing, and increasing the priority of the backup or relay path; the second threshold is a preset path low stability score threshold.

7. The method according to claim 1, characterized in that, The dynamic adjustment of the number of concurrent paths and the racing window in the routing control parameters based on the path scoring results includes: When a candidate path is detected to have a preset potential risk, a race suppression action is executed to achieve race suppression and transmission concurrency peak reduction. The preset potential risks include increased fluctuations in path round-trip delay, decreased availability of candidate relay paths, tightening of budget status, or path risk score exceeding a preset safety threshold. The race suppression action includes reducing the number of concurrent paths in this round, tightening the time window for race selection, increasing the frequency of backup path detection, and reducing the participation priority of candidate paths identified as high-risk in this round of race.

8. A network path scoring and adaptive routing system, characterized in that, The system includes a processor and a memory; the memory stores a computer program; the processor is configured to execute the computer program to implement the overlay network path scoring and adaptive routing method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, The electronic device includes at least one processor and a memory communicating with the at least one processor. The memory stores a computer program that, when executed by the processor, causes the electronic device to implement the overlay network path scoring and adaptive routing method according to any one of claims 1 to 7. The electronic device also includes at least one communication interface for interacting with the control plane and data plane of the overlay network. The processor receives service semantic features issued by the control plane through the communication interface and obtains path status features of the candidate path set through a local network module.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by instructions of a processor, causes the processor to implement the overlay network path scoring and adaptive routing method according to any one of claims 1 to 7.

Images