Distributed large-scale layer 2 network intelligent routing method and system based on edge cloud nodes

Abstract

The present invention relates to a distributed large-scale Layer 2 network intelligent routing method and system based on edge cloud nodes, and belongs to the field of computer network technology. In view of the problems of slow response to routing decisions and low fault recovery efficiency in existing networks under dynamic environments, it is proposed to periodically collect link bandwidth utilization, transmission delay and other status information through edge cloud nodes, upload them to the central cloud platform to generate a global routing strategy and send it down to the edge nodes, and trigger fault switching in combination with local dynamic threshold monitoring. The technical solution includes two-way communication between edge nodes and the central platform, a neural network model for spatiotemporal feature fusion, and a multi-objective optimized path priority generation mechanism. The present method and system are suitable for large-scale distributed network environments and can improve network resource utilization and business continuity.

Description

Technical Field

[0001] The present invention relates to the field of computer network technology. More specifically, the present invention relates to a distributed large-scale Layer 2 network intelligent routing method and system based on edge cloud nodes. Background Art

[0002] In existing distributed network environments, traditional routing protocols (such as OSPF and BGP) rely on static configuration or centralized control, making them difficult to adapt to dynamically changing network conditions. Issues such as link bandwidth fluctuations between edge nodes, sudden increases in transmission latency, and uneven node load often lead to lagging routing policies, causing business interruptions or decreased service quality. For example, in industrial IoT scenarios, frequent device access and bursty data transmissions can easily cause localized congestion, while centralized routing calculations cannot respond promptly due to data transmission delays. Furthermore, existing failover mechanisms are often based on fixed thresholds and lack the ability to predict latency fluctuations, resulting in a high failure rate for switchover path verification.

[0003] The reasons for these issues include: 1) Global network status awareness relies on periodic reporting, resulting in insufficient real-time data; 2) Routing policy optimization fails to integrate spatiotemporal features, making it difficult to balance multi-dimensional constraints (bandwidth, latency, and load); and 3) Local decision-making lacks an adaptive threshold adjustment mechanism, making it incapable of responding to unexpected anomalies. Existing solutions attempt to incorporate SDN or AI models, but they still face challenges such as limited edge computing resources, high model training costs, and inefficient cross-platform policy synchronization. Summary of the Invention

[0004] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0005] In order to achieve these purposes and other advantages according to the present invention, a distributed large-scale Layer 2 network intelligent routing method based on edge cloud nodes is provided, comprising the following steps:

[0006] Multiple edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol. Each edge cloud node periodically collects network status information of the node where it is located. The network status information includes link bandwidth utilization between adjacent nodes, cross-node transmission delay, network device port status, and node local computing resource load rate;

[0007] Each edge cloud node encapsulates the network status information into a structured data packet and uploads it to the central cloud platform via an encrypted tunnel;

[0008] Based on the network status information of all edge cloud nodes, the central cloud platform uses a neural network model based on spatiotemporal feature extraction to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths between edge cloud nodes.

[0009] After receiving the global routing policy issued by the central cloud platform, each edge cloud node updates the forwarding table entries in the local routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding table entries to the adjacent nodes through the pre-configured communication protocol;

[0010] Each edge cloud node monitors the link transmission delay variation between adjacent nodes in real time. When it detects that the link transmission delay variation exceeds the dynamic threshold pre-stored in the local routing decision module, it triggers the local routing decision module to reconfigure the local routing table based on the failover path set in the global routing policy, and marks the link status anomaly information and uploads it to the central cloud platform via an encrypted tunnel.

[0011] The central cloud platform initiates the recalculation of the global routing policy based on the link status anomaly information, and distributes the updated global routing policy to the affected edge cloud nodes associated with the link status anomaly information.

[0012] Preferably, the neural network model based on spatiotemporal feature extraction includes a convolutional neural network branch and a long short-term memory network branch arranged in parallel, wherein:

[0013] The convolutional neural network branch uses the physical topological connection relationship of edge cloud nodes as input and extracts spatial correlation features between adjacent nodes through three convolutional layers. The spatial correlation features include the link bandwidth fluctuation pattern and port status change trend between nodes.

[0014] The long short-term memory network branch takes the cross-node transmission delay sequence as input and extracts the delay fluctuation time series characteristics within a preset time window through a bidirectional loop structure. The delay fluctuation time series characteristics include periodic congestion patterns and sudden abnormal delay segments;

[0015] The neural network model dynamically assigns weights to spatial correlation features and delay fluctuation time series features through the attention mechanism layer to generate a network state vector that integrates spatiotemporal features.

[0016] The fully connected layer of the neural network model generates a global routing strategy based on the network state vector. The generation process of the path priority list includes a multi-objective optimization operation on the network state vector that integrates spatiotemporal features. The constraints of the multi-objective optimization operation include the link bandwidth utilization threshold, the transmission delay upper limit, and the node load balancing coefficient.

[0017] Preferably, reconfiguring the local routing table storage unit based on the failover path set includes the following steps:

[0018] When it is detected that the link transmission delay variation exceeds the dynamic threshold, a pre-sorted candidate path sequence is extracted from the failover path set, and the candidate path sequence is prioritized based on the historical transmission success rate and the number of path hops;

[0019] The highest-priority candidate path is tested for link bandwidth availability and transmission delay. Link bandwidth availability is tested by sending probe packets to the target node and calculating the packet loss rate. Transmission delay is verified by measuring the difference between the round-trip delay of the probe packets and a preset delay threshold.

[0020] If the packet loss rate of the candidate path is lower than the preset packet loss threshold and the round-trip delay difference is within the preset tolerance range, the traffic of the current faulty link is diverted to the candidate path according to the preset ratio, and a forwarding table entry update instruction including the diversion ratio and path identifier is generated;

[0021] If the candidate path verification fails, the dynamic weight adjustment of the candidate path sequence is triggered, the path priority is recalculated according to the real-time load rate of the adjacent nodes, and the link bandwidth availability detection step is jumped to perform the suboptimal candidate path verification;

[0022] After the path switch is completed, the local routing decision module stores the updated forwarding table entry in association with the failover timestamp and sends a routing table synchronization request to the adjacent edge cloud node through a preconfigured communication protocol. The synchronization request includes the changed forwarding table entry hash value and version number.

[0023] The network status detection unit continuously monitors the transmission quality indicators of the switched path. When the transmission delay fluctuation exceeds 50% of the dynamic threshold within three consecutive detection cycles, the secondary path switching process is triggered and the priority weight parameters of the failover path set are updated.

[0024] Preferably, the execution process of the multi-objective optimization operation includes the following steps:

[0025] The network state vector that integrates spatiotemporal features is mapped to a bandwidth allocation scaling factor α i , delay compensation factor β j and node load balancing weight γ k A set of decision variables, where i corresponds to the path number, j corresponds to the delay level identifier, and k corresponds to the node identifier;

[0026] Construct an objective function set including the first objective function, the second objective function and the third objective function. The first objective function is f1=Σ(α i ×C i ), used to maximize the effective bandwidth utilization, the second objective function is f2=Σ(β j ×D j), used to minimize the transmission delay offset, the third objective function is f3=Σ(γ k ×L k ), used to balance the node load differences, where C i represents the available bandwidth capacity of path i, D j represents the delay compensation amount of path j, L k represents the load deviation value of node k;

[0027] Set up a system of equations with the following constraints:

[0028] Σα i ≤ predefined link bandwidth utilization threshold, Σβ j ≤predefined transmission delay upper limit, Σγ k ≤predefined node load balancing coefficient;

[0029] A non-dominated sorting genetic algorithm with an elite retention strategy is used to solve the objective function set, which includes:

[0030] An initial solution set is generated based on the path priority list as the genetic algorithm population. An adaptive crossover operation based on path similarity is performed. The crossover probability is dynamically adjusted according to the path overlap rate of the individuals in the solution set. An adjacent path replacement mutation operation is performed to replace the current path with a candidate path that is directly adjacent in the physical topology.

[0031] The following processing is performed on the solution set produced by each iteration:

[0032] Delete the individuals that violate the constraint equation group, perform non-dominated sorting on the remaining individuals and calculate the crowding distance, and retain the first preset number of individuals with the highest sorting level and the largest crowding distance;

[0033] From the Pareto frontier solution set of the final iterative result, select the individual with the largest crowding distance as the optimal solution, and extract the α corresponding to the individual. i , β j and γ k The parameters are combined to generate a ranking weight value of the path priority list.

[0034] Preferably, the process of determining and updating the dynamic threshold in the local routing decision module includes the following steps:

[0035] The local routing decision module loads the baseline threshold T0 when it is initialized. The baseline threshold T0 is calculated based on the normal distribution parameters of historical transmission delay data. The calculation formula is T0 = μ + 3σ, where μ is the mean delay and σ is the standard deviation of the delay.

[0036] The network status detection unit calculates the sliding window average of the link transmission delay between adjacent nodes every five minutes. The window size is the most recent 30 sampling points. When the deviation between the sliding window average and the baseline threshold T0 exceeds 20%, the threshold adaptive adjustment is triggered.

[0037] The threshold adaptive adjustment process performs the following operations:

[0038] Collect the latency data set for each hour within 24 hours before the current time, remove data points with a value exceeding three times the standard deviation, recalculate the mean μ' and standard deviation σ', generate the updated dynamic threshold T1 = μ' + 2.5σ', and overwrite and store it in the local routing decision module;

[0039] If a single delay change is detected to exceed 150% of the current dynamic threshold between two threshold adjustment cycles, the emergency threshold correction process is immediately initiated, temporarily setting the current dynamic threshold to 80% of the abnormal delay value and marking it as pending calibration.

[0040] When the edge cloud node receives the updated global routing policy issued by the central cloud platform, it synchronously obtains the threshold calibration parameters contained in the global routing policy, and fuses the local dynamic threshold with the calibration parameters through the weighted average algorithm to generate the final dynamic threshold.

[0041] Preferably, the process of periodically collecting network status information includes the following steps:

[0042] The network status detection unit configures multi-dimensional data collection tasks according to the preset collection strategy. The collection strategy includes a 5-minute collection period for link bandwidth utilization, a 30-second collection period for cross-node transmission delay, a 1-minute collection period for network device port status, and a 2-minute collection period for node local computing resource load rate.

[0043] When collecting link bandwidth utilization, ICMP probe packets are sent to adjacent nodes and the arithmetic average of the bandwidth utilization of the last 10 probe results is calculated. The real-time throughput counter value of the network device port is also obtained through the SNMP protocol.

[0044] When performing cross-node transmission delay collection, a UDP benchmark test package is used for bidirectional transmission testing. The average time difference between the sending time and the receiving confirmation time is calculated as the effective delay value, and the maximum delay fluctuation during the test is recorded.

[0045] The collected raw data is processed by the preprocessing module to perform the following operations:

[0046] Abnormal sampling points exceeding three times the standard deviation are eliminated, the binary flag bits in the port status data are converted into one-hot encoding, and indicator data of different dimensions are converted into feature values ​​of uniform magnitude using the Z-score normalization method;

[0047] When the network status detection unit detects that the change rate of two consecutive collection results of any indicator exceeds the preset sensitivity threshold, it automatically shortens the indicator collection cycle to 50% of the original cycle and starts intensive sampling for three consecutive cycles;

[0048] All pre-processed network status information data is attached with a timestamp accurate to the millisecond level when stored, and the clock is synchronized and calibrated with the central cloud platform through the NTP protocol.

[0049] Preferably, the processing after the suboptimal candidate path verification fails further includes the following steps:

[0050] If the candidate path fails verification three times in a row, the local routing decision module performs the following operations:

[0051] Roll back the traffic on the current faulty link to the path before the switchover and generate a path rollback instruction to update the forwarding table entry.

[0052] Send a path switching failure alarm to the central cloud platform. The alarm includes the failed path identifier, verification failure reason code, and current node load status.

[0053] Activate local degradation mode to limit the forwarding rate of non-critical business traffic to a preset safety threshold while maintaining the minimum guaranteed bandwidth for critical business traffic;

[0054] After receiving the path switching failure alarm information, the central cloud platform marks the path as a high-risk path in the updated global routing policy and excludes it from the failover path set for at least 24 hours.

[0055] Provides a distributed large-scale Layer 2 network intelligent routing system based on edge cloud nodes, including:

[0056] Multiple edge cloud nodes and central cloud platforms. The edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol;

[0057] Each edge cloud node includes a network status detection unit, a local routing decision module, and a routing table storage unit. The network status detection unit periodically collects link bandwidth utilization, cross-node transmission delay, network device port status, and node local computing resource load rate between adjacent nodes;

[0058] The local routing decision module encapsulates the data collected by the network status detection unit into structured data packets and transmits them to the central cloud platform through an encrypted tunnel;

[0059] The central cloud platform is equipped with a global routing analysis engine. Based on the network status information received from all edge cloud nodes, the global routing analysis engine uses a spatiotemporal feature extraction neural network model to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths.

[0060] After receiving the global routing policy issued by the central cloud platform, the local routing decision module updates the forwarding entries in the routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding entries to the adjacent edge cloud nodes through the pre-configured communication protocol;

[0061] The network status detection unit monitors the variation of link transmission delay between adjacent nodes in real time. When it detects that the variation exceeds the dynamic threshold pre-stored in the routing table storage unit, it triggers the local routing decision module to reconfigure the routing table storage unit based on the failover path set and upload the marked link status anomaly information to the central cloud platform via an encrypted tunnel.

[0062] The global routing analysis engine of the central cloud platform initiates policy recalculation based on the link status anomaly information and distributes the updated global routing policy to the associated affected edge cloud nodes.

[0063] The present invention has at least the following beneficial effects:

[0064] First, by periodically collecting network status information (such as link bandwidth utilization and transmission latency) through edge cloud nodes and combining it with dynamic threshold monitoring, the system can quickly detect changes in network status. The central cloud platform generates global routing policies in real time based on a neural network model that integrates spatiotemporal features. This effectively addresses sudden traffic fluctuations and uneven node loads, mitigating policy failures caused by data lags in traditional routing protocols.

[0065] Second, a distributed architecture enables edge nodes to perform local decisions (such as failover path verification) to reduce computational pressure on the central platform. Simultaneously, the central platform provides a global path priority list using a multi-objective optimization algorithm. These two elements work together to ensure real-time local response while balancing optimal allocation of network-wide resources, avoiding the delays and resource bottlenecks associated with a single centralized control system.

[0066] Third, by combining dynamic threshold triggering (e.g., when latency variation exceeds a preset threshold) with multi-path verification (packet loss rate detection and latency tolerance verification), the system can rapidly switch to pre-ranked candidate paths. Local degradation mode restricts non-critical service traffic in the event of repeated switch failures, prioritizing core services. This significantly reduces service interruption and improves network fault tolerance.

[0067] Fourth, a multi-objective optimization algorithm balances multiple constraints, including bandwidth utilization, transmission latency, and node load balancing, to maximize network resource efficiency. Encrypted tunnel communication (such as the TLS protocol) and a policy signature mechanism ensure the security of data transmission and policy synchronization, preventing malicious attacks and data tampering, and enhancing overall system stability.

[0068] Fifth, dynamic threshold adjustment mechanisms (e.g., based on sliding window statistics and global calibration parameter fusion) enable the system to adapt to changes in diverse network environments. The modular design of edge nodes and the central platform supports flexible expansion and is suitable for a variety of large-scale distributed network scenarios, such as the Industrial Internet of Things and cloud computing centers.

[0069] Other advantages, objectives and features of the present invention will be reflected in part from the following description and will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a schematic diagram of the network status collection and upload process of the present invention;

[0071] Figure 2 This is a structural diagram of the spatiotemporal feature fusion neural network model of the present invention. DETAILED DESCRIPTION

[0072] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0073] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified; in the description of the present invention, the orientation or positional relationship indicated by the terms is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0074] like Figure 1~Figure 2 As shown, the present invention provides a distributed large-scale Layer 2 network intelligent routing method based on edge cloud nodes, comprising the following steps:

[0075] Multiple edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol. Each edge cloud node periodically collects network status information of the node where it is located. The network status information includes link bandwidth utilization between adjacent nodes, cross-node transmission delay, network device port status, and node local computing resource load rate;

[0076] Each edge cloud node encapsulates the network status information into a structured data packet and uploads it to the central cloud platform via an encrypted tunnel;

[0077] Based on the network status information of all edge cloud nodes, the central cloud platform uses a neural network model based on spatiotemporal feature extraction to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths between edge cloud nodes.

[0078] After receiving the global routing policy issued by the central cloud platform, each edge cloud node updates the forwarding table entries in the local routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding table entries to the adjacent nodes through the pre-configured communication protocol;

[0079] Each edge cloud node monitors the link transmission delay variation between adjacent nodes in real time. When it detects that the link transmission delay variation exceeds the dynamic threshold pre-stored in the local routing decision module, it triggers the local routing decision module to reconfigure the local routing table based on the failover path set in the global routing policy, and marks the link status anomaly information and uploads it to the central cloud platform via an encrypted tunnel.

[0080] The central cloud platform initiates the recalculation of the global routing policy based on the link status anomaly information, and distributes the updated global routing policy to the affected edge cloud nodes associated with the link status anomaly information.

[0081] In the above technical solution, edge cloud nodes periodically collect network status information including link bandwidth utilization, inter-node transmission delay, network device port status, and node local computing resource load. The collection period can be set to every 5 minutes for link bandwidth utilization, every 30 seconds for inter-node transmission delay, every 1 minute for port status, and every 2 minutes for computing resource load. Bandwidth utilization can be determined by sending ICMP probe packets and calculating the arithmetic mean of the last 10 probe results. The real-time port throughput counter value is also obtained via the SNMP protocol. Inter-node transmission delay is tested using a UDP benchmark test packet for bidirectional transmission, and the average time difference is calculated as the effective delay value.

[0082] In terms of equipment selection, edge cloud nodes can utilize servers with multi-core processors, network switches supporting SNMP and ICMP protocols, and embedded gateways supporting UDP testing. Regarding materials, the server chassis can be constructed of aluminum alloy, with internal circuit boards constructed of FR-4 epoxy resin. Edge cloud nodes are typically deployed at the edge of the network, close to end devices, such as at base stations or within factory floors.

[0083] During operation, the network status detection unit performs data collection tasks according to the preset cycle. After the pre-processing module eliminates abnormal sampling points, it adds millisecond-level timestamps to the standardized data and uploads it to the central cloud platform through an encrypted tunnel.

[0084] Global routing policy generation and issuance: The central cloud platform uses a neural network model based on spatiotemporal feature extraction to generate global routing policies. The convolutional neural network branch inputs physical topological connectivity relationships and extracts spatial correlation features through three convolutional layers. The long-short-term memory network branch inputs delay sequences and extracts delay fluctuation characteristics. The attention mechanism layer dynamically assigns weights to the two types of features, and the fully connected layer outputs a list of path priorities, traffic distribution weights, and a set of failover paths.

[0085] In terms of equipment selection, the central cloud platform can utilize a high-performance server cluster supporting GPU acceleration, and neural network models can be implemented based on the TensorFlow or PyTorch frameworks. Regarding materials, the server cooling module can utilize a combination of copper heat pipes and aluminum alloy fins. The central cloud platform is deployed in the data center computer room and connected to the edge cloud nodes via redundant fiber optic links.

[0086] During operation, the global routing analysis engine receives data from all edge cloud nodes, generates a strategy using a multi-objective optimization algorithm, and encapsulates the strategy in JSON format for distribution. The path priority list is sorted using a non-dominated sorting genetic algorithm, subject to constraints such as link bandwidth utilization thresholds (e.g., 85%), transmission latency caps (e.g., 100ms), and node load balancing coefficients (e.g., 0.3).

[0087] Local routing table update and failover: The edge cloud node monitors the variation in link transmission delay, with a dynamic threshold initialized to T0 = μ + 3σ (for example, T0 = 80ms when μ = 50ms and σ = 10ms). When the delay variation exceeds the threshold, a failover process is triggered. Candidate path verification includes packet loss rate detection (with a preset packet loss threshold of 2%) and round-trip delay verification (with a preset tolerance of ±10ms). If verification passes, traffic on the faulty link is diverted to the candidate path at a preset ratio (for example, 70%). If verification fails, path priority is dynamically adjusted and a suboptimal path is attempted.

[0088] Regarding device selection, the local routing decision module can utilize a software-defined networking (SDN) controller that supports the OpenFlow protocol, and the routing table storage unit can utilize a high-speed NVMe solid-state drive. Regarding materials, the device casing can be constructed from flame-retardant ABS plastic. Edge cloud nodes must maintain direct physical connectivity with adjacent nodes, such as via fiber optic or Cat6a cables.

[0089] During operation, the local routing decision module receives the policy and updates the forwarding table entries, synchronizing the changes to neighboring nodes via BGP. After a path switchover, the network status detection unit continuously monitors transmission quality. If latency fluctuation exceeds 50% of a dynamic threshold (e.g., 40ms), a secondary switchover is triggered and the failover path set is updated.

[0090] This method achieves real-time network status awareness and global optimization through periodic data collection and dynamic policy adjustment, reducing service interruptions caused by link congestion or node overload. A failover mechanism combined with multipath verification improves network fault tolerance. A neural network model that integrates spatiotemporal features enhances the adaptability and accuracy of routing decisions, ultimately improving the stability and transmission efficiency of large-scale Layer 2 networks.

[0091] In another technical solution, the neural network model based on spatiotemporal feature extraction includes a convolutional neural network branch and a long short-term memory network branch arranged in parallel, wherein:

[0092] The convolutional neural network branch uses the physical topological connection relationship of edge cloud nodes as input and extracts spatial correlation features between adjacent nodes through three convolutional layers. The spatial correlation features include the link bandwidth fluctuation pattern and port status change trend between nodes.

[0093] The long short-term memory network branch takes the cross-node transmission delay sequence as input and extracts the delay fluctuation time series characteristics within a preset time window through a bidirectional loop structure. The delay fluctuation time series characteristics include periodic congestion patterns and sudden abnormal delay segments;

[0094] The neural network model dynamically assigns weights to spatial correlation features and delay fluctuation time series features through the attention mechanism layer to generate a network state vector that integrates spatiotemporal features.

[0095] The fully connected layer of the neural network model generates a global routing strategy based on the network state vector. The generation process of the path priority list includes a multi-objective optimization operation on the network state vector that integrates spatiotemporal features. The constraints of the multi-objective optimization operation include the link bandwidth utilization threshold, the transmission delay upper limit, and the node load balancing coefficient.

[0096] In the above technical solution, the convolutional neural network branch inputs the physical topological connectivity of edge cloud nodes, using a three-layer convolutional layer structure. The number of convolution kernels in each layer can be 32, 64, or 128, respectively, and the kernel size can be set to 3×3. Spatial correlation features extracted include link bandwidth fluctuation patterns (e.g., bandwidth change rate within 5 minutes) and port status change trends (e.g., port enable / disable frequency). The long short-term memory network branch inputs the cross-node transmission delay sequence, with a time window set to 60 seconds and a configurable number of hidden layer units of 128. Temporal features extracted include periodic congestion patterns (e.g., hourly peak delay) and sudden abnormal delay segments (e.g., segments with delay surges exceeding 50ms).

[0097] Regarding device selection, the neural network model can be deployed on a GPU server supporting CUDA acceleration, and the convolutional and LSTM layers can be implemented using the TensorFlow framework. Regarding materials, the server motherboard can be made of a glass fiber-reinforced epoxy resin substrate, and the heat sink can be made of aluminum alloy. The neural network module must be deployed in a high-performance computing node on the central cloud platform and connected to the data storage unit via a PCIe bus.

[0098] During operation, physical topology data is fed into the CNN branch in the form of an adjacency matrix, while time delay sequence data is fed into the LSTM branch by time step. The CNN branch extracts spatial correlation features between nodes through convolutional layers, while the LSTM branch captures delay fluctuation trends through a bidirectional recurrent structure. The output feature vectors are then passed to the attention mechanism layer.

[0099] The attention mechanism layer dynamically assigns weights to the spatial correlation features output by the CNN and the time-delay fluctuation temporal features output by the LSTM. Weight assignment can be calculated based on the cosine similarity of the feature vectors. For example, the weight coefficient for spatial features can be set to 0.6, and the weight coefficient for temporal features can be set to 0.4. The fused network state vector can be set to 256 dimensions and mapped to a low-dimensional feature representation through a fully connected layer.

[0100] Regarding equipment selection, the attention mechanism module can be implemented based on open-source deep learning libraries (such as PyTorch's MultiheadAttention layer). Regarding materials, copper wires can be used for the data bus, and RJ45 connectors with gold-plated contacts can be used for the interface connector. This module needs to be integrated into the global routing analysis engine of the central cloud platform and interact with the neural network branches via high-speed memory channels.

[0101] During operation, the attention mechanism calculates the correlation scores between spatial and temporal features and weights the feature vectors proportionally. The fused network state vector is normalized through a fully connected layer and output to the multi-objective optimization module. During the feature fusion phase, timestamp alignment must be ensured to avoid feature misalignment due to differences in data collection cycles.

[0102] The decision variables for the multi-objective optimization operation include the bandwidth allocation factor α (e.g., α1 = 0.7, α2 = 0.3), the delay compensation factor β (e.g., β1 = 1.2, β2 = 0.8), and the node load balancing weight γ (e.g., γ1 = 0.5, γ2 = 0.5). Constraints include a link bandwidth utilization threshold (85%), an upper bound on transmission delay (100ms), and a node load balancing coefficient (0.3). The optimization algorithm uses a non-dominated sorting genetic algorithm (NSGA-II) with an elitist retention strategy. The population size can be set to 200, and the number of iterations is 100.

[0103] Regarding equipment selection, the multi-objective optimization module can run on servers equipped with multi-core CPUs, and the genetic algorithm's crossover and mutation operations can be implemented based on the DEAP framework. Regarding materials, the server memory modules can use DDR4 DRAM chips, and the storage units can be equipped with TLC NAND flash memory. This module should be deployed in the policy computing node of the central cloud platform, exchanging data with the neural network module through shared memory.

[0104] During operation, the initial solution set is generated based on historical routing strategies. The crossover operation uses a two-point crossover method, with the crossover probability dynamically adjusted based on the path overlap ratio (for example, the probability drops to 0.3 when the overlap ratio exceeds 80%). The mutation operation replaces the current path with an adjacent candidate path in the physical topology (for example, from path A to path B). The Pareto frontier solution set is filtered by congestion distance, and the individual with the largest congestion distance is ultimately selected to form the path priority list.

[0105] This method uses CNN and LSTM branches to extract spatial topological features and latency fluctuation patterns, respectively, and combines this with a dynamic weight fusion mechanism to enhance the adaptability of routing strategies to network conditions. A multi-objective optimization algorithm generates a globally optimal path under multiple constraints, including bandwidth, latency, and load balancing, mitigating the risk of link congestion. The synergy between the attention mechanism and the genetic algorithm improves routing decision efficiency and is suitable for real-time traffic scheduling in large-scale edge network environments.

[0106] In another technical solution, reconfiguring the local routing table storage unit based on the failover path set includes the following steps:

[0107] When it is detected that the link transmission delay variation exceeds the dynamic threshold, a pre-sorted candidate path sequence is extracted from the failover path set, and the candidate path sequence is prioritized based on the historical transmission success rate and the number of path hops;

[0108] The highest-priority candidate path is tested for link bandwidth availability and transmission delay. Link bandwidth availability is tested by sending probe packets to the target node and calculating the packet loss rate. Transmission delay is verified by measuring the difference between the round-trip delay of the probe packets and a preset delay threshold.

[0109] If the packet loss rate of the candidate path is lower than the preset packet loss threshold and the round-trip delay difference is within the preset tolerance range, the traffic of the current faulty link is diverted to the candidate path according to the preset ratio, and a forwarding table entry update instruction including the diversion ratio and path identifier is generated;

[0110] If the candidate path verification fails, the dynamic weight adjustment of the candidate path sequence is triggered, the path priority is recalculated according to the real-time load rate of the adjacent nodes, and the link bandwidth availability detection step is jumped to perform the suboptimal candidate path verification;

[0111] After the path switch is completed, the local routing decision module stores the updated forwarding table entry in association with the failover timestamp and sends a routing table synchronization request to the adjacent edge cloud node through a preconfigured communication protocol. The synchronization request includes the changed forwarding table entry hash value and version number.

[0112] The network status detection unit continuously monitors the transmission quality indicators of the switched path. When the transmission delay fluctuation exceeds 50% of the dynamic threshold within three consecutive detection cycles, the secondary path switching process is triggered and the priority weight parameters of the failover path set are updated.

[0113] In the above technical solution, when the link transmission delay variation is detected to exceed a dynamic threshold (e.g., 80ms), a pre-ordered sequence of candidate paths is extracted from the set of failover paths. Candidate paths are prioritized based on their historical transmission success rate (e.g., paths with a success rate ≥ 95% are prioritized) and path hop count (e.g., paths with a hop count ≤ 3 are prioritized). The historical transmission success rate is calculated by counting the path's packet loss rate over the past 24 hours, and the path hop count is obtained in real time from the network topology database.

[0114] In terms of device selection, the failover path set can be stored in an in-memory database supporting the Redis protocol, and the path priority calculation module can be implemented on an embedded processor with multi-threaded processing capabilities. Regarding materials, the in-memory database's memory chip can use LPDDR4 specifications, and the processor package substrate can be made of FR-4 epoxy resin. This module should be deployed in the local routing decision module of the edge cloud node and connected to the network status detection unit via an internal bus.

[0115] During operation, the local routing decision module triggers candidate path extraction based on dynamic thresholds, prioritizing paths with high historical transmission success rates and few hops. The candidate path sequence is stored in the routing table storage unit in JSON format, and a hash algorithm is used to ensure data integrity.

[0116] The system checks the link bandwidth availability of the highest-priority candidate path, sending probe packets to calculate the packet loss rate (with a preset loss threshold of 2%) and measure the round-trip latency (with a preset tolerance of ±10ms). If the verification passes (for example, a packet loss rate of 1.5% and a latency difference of 8ms), traffic from the faulty link is diverted to that path at a preset ratio (for example, 70%). A forwarding table update is generated, containing the diversion ratio and the path ID.

[0117] Regarding equipment selection, detection packets can be sent using hardware probes that support the IP SLA protocol, and latency measurement can be performed using a high-precision clock synchronization chip. Regarding materials, the probe housing can be made of flame-retardant ABS plastic, and the clock chip package can be encapsulated using a ceramic substrate. The detection module needs to be integrated into the network interface card of the edge cloud node and communicate with the routing decision module via the PCIe interface.

[0118] During operation, the verification module sends UDP probe packets to the target node, calculates the packet loss rate, and calculates the average latency. If the result meets the threshold, the routing table storage unit updates the forwarding entry and synchronizes the change information with neighboring nodes via the BGP protocol. The traffic diversion ratio is dynamically adjusted based on the remaining bandwidth of the path. For example, when the remaining bandwidth is ≥50Mbps, 70% of the traffic is diverted.

[0119] If a candidate path fails verification (for example, a packet loss rate of 3% or a latency difference of 15ms), dynamic path priority adjustment is triggered. The path priority is recalculated based on the real-time load of adjacent nodes (for example, nodes with a load rate ≥ 80% are weighted lower), and the next-best candidate path is tried. If verification fails three times in a row, traffic on the failed link is reverted to the original path, and the forwarding rate of non-critical traffic is limited to a preset safety threshold (for example, 10Mbps).

[0120] Regarding equipment selection, load rate collection can be based on a network traffic analyzer that supports the NetFlow protocol, and rate limiting can be implemented using a switch chip that supports QoS policies. Regarding materials, the analyzer housing can be made of aluminum alloy, and the heat sink can be made of copper heat pipes. This module should be deployed in the traffic control unit of the edge cloud node and interact with the routing decision module through shared memory.

[0121] During operation, the dynamic adjustment module reorders candidate paths based on node load data and triggers suboptimal path verification. During a fallback operation, the forwarding table entries are restored to their pre-switching state, and an alarm is sent to the central cloud platform via the SNMP protocol. The network status detection unit continuously monitors the transmission quality of the post-switching path. If the latency fluctuation exceeds 50% of the dynamic threshold (e.g., 40ms) for three consecutive cycles, a secondary path switch is triggered and the priority weights of the failover path set are updated.

[0122] This approach ensures reliable and timely failover through pre-sorted candidate paths and a multi-dimensional verification mechanism. Dynamic priority adjustment prevents secondary congestion caused by node overload, while fallback mechanisms and traffic limiting minimize service disruptions. Continuous monitoring and secondary switching further optimize path stability, making it suitable for rapid recovery in highly dynamic network environments.

[0123] In another technical solution, the execution process of the multi-objective optimization operation includes the following steps:

[0124] The network state vector that integrates spatiotemporal features is mapped to a bandwidth allocation scaling factor α i , delay compensation factor β j and node load balancing weight γ k A set of decision variables, where i corresponds to the path number, j corresponds to the delay level identifier, and k corresponds to the node identifier;

[0125] Construct an objective function set including the first objective function, the second objective function and the third objective function. The first objective function is f1=Σ(α i ×C i ), used to maximize the effective bandwidth utilization, the second objective function is f2=Σ(β j ×D j ), used to minimize the transmission delay offset, the third objective function is f3=Σ(γ k ×L k ), used to balance the node load differences, where C i represents the available bandwidth capacity of path i, D j represents the delay compensation amount of path j, L k represents the load deviation value of node k;

[0126] Set up a system of equations with the following constraints:

[0127] Σα i ≤ predefined link bandwidth utilization threshold, Σβ j ≤predefined transmission delay upper limit, Σγ k ≤predefined node load balancing coefficient;

[0128] A non-dominated sorting genetic algorithm with an elite retention strategy is used to solve the objective function set, which includes:

[0129] An initial solution set is generated based on the path priority list as the genetic algorithm population. An adaptive crossover operation based on path similarity is performed. The crossover probability is dynamically adjusted according to the path overlap rate of the individuals in the solution set. An adjacent path replacement mutation operation is performed to replace the current path with a candidate path that is directly adjacent in the physical topology.

[0130] The following processing is performed on the solution set produced by each iteration:

[0131] Delete the individuals that violate the constraint equation group, perform non-dominated sorting on the remaining individuals and calculate the crowding distance, and retain the first preset number of individuals with the highest sorting level and the largest crowding distance;

[0132] From the Pareto frontier solution set of the final iterative result, select the individual with the largest crowding distance as the optimal solution, and extract the α corresponding to the individual. i , β j and γ k The parameters are combined to generate a ranking weight value of the path priority list.

[0133] In the above technical solution, the network state vector integrating spatiotemporal features is mapped to the bandwidth allocation factor α (e.g., α1=0.6, α2=0.4), the delay compensation factor β (e.g., β1=1.1, β2=0.9), and the node load balancing weight γ (e.g., γ1=0.5, γ2=0.5). The objective function includes maximizing the effective bandwidth utilization (f1=Σα i ×C i , C i is the available bandwidth capacity of the path), minimize the transmission delay offset (f2=Σ(β j ×D j ), D j is the path delay compensation amount) and the balance node load difference (f3=Σ(γ k ×L k ), L k is the node load deviation value).

[0134] Regarding device selection, the mapping calculation can be run on a multi-core CPU server supporting floating-point acceleration, and the objective function calculation module can be implemented using the NumPy library. Regarding materials, the server motherboard can be constructed using an FR-4 epoxy resin substrate, and the memory module can be equipped with DDR4 DRAM chips. This module must be deployed in the policy computing node of the central cloud platform and connected to the neural network module via a high-speed data bus.

[0135] During operation, the network state vector is normalized by the fully connected layer and then fed into the variable mapping module, which generates a set of decision variables based on predefined weighting rules. The objective function calculates the bandwidth, latency, and load indicators of each path in real time, and performs multi-objective correlation analysis through matrix operations.

[0136] Constraints include a link bandwidth utilization threshold (85%), a transmission latency cap (100ms), and a node load balancing coefficient (0.3). The optimization algorithm uses a non-dominated sorting genetic algorithm (NSGA-II) with an elitist retention strategy. The population size can be set to 200, and the number of iterations is 100. The crossover operation uses a two-point crossover method, with the crossover probability dynamically adjusted based on the path overlap ratio (for example, the probability is reduced to 0.3 when the overlap ratio is ≥80%). The mutation operation involves adjacent path replacement (for example, replacing path A with the topologically adjacent path B).

[0137] Regarding equipment selection, the genetic algorithm solver can be run on a multi-threaded parallel computing cluster, and crossover and mutation operations can be implemented based on the DEAP framework. Regarding materials, cluster nodes can use copper heat sinks and aluminum alloy chassis, and storage units can be equipped with TLC NAND flash memory. This module must be integrated into the global routing analysis engine of the central cloud platform and interact with the objective function module through shared memory.

[0138] During operation, the initial solution set is generated based on the historical routing strategy. Each generation of the population generates new individuals through crossover and mutation, and solutions that violate the constraints are deleted. The remaining individuals are sorted by non-dominated hierarchy, and the crowding distance is used to prioritize within the same hierarchy. Ultimately, the top 50 best individuals are retained for the next iteration.

[0139] The Pareto frontier solution set from the final iteration is calculated for its crowding distance, and the individual with the largest distance is selected as the optimal solution (for example, an individual with a crowding distance ≥ 0.8). The optimal solution's α, β, and γ parameter combinations are extracted and converted into ranking weights for the path priority list (for example, α1 = 0.6 corresponds to a weight of 60% for path 1). The path priority list is distributed to the edge cloud node in JSON format.

[0140] Regarding device selection, the congestion distance calculation can be based on a mathematical acceleration library that supports vectorized operations, and the parameter extraction module can be implemented on a processor with a high-precision floating-point unit. Regarding materials, the processor package can be made of a ceramic substrate, and the data interface can be a PCIe slot with gold-plated contacts. This module should be deployed in the policy distribution node of the central cloud platform and connected to the genetic algorithm module via a high-speed network.

[0141] During operation, the Pareto solution set is sorted by non-dominated sorting, and the congestion distance of each solution is calculated. The optimal solution parameters are converted into path weights through linear interpolation, and executable routing policy instructions are generated. The policy instructions are encrypted and distributed to the associated edge cloud nodes via fiber links.

[0142] This method uses a multi-objective optimization algorithm to balance multiple constraints, including bandwidth utilization, transmission latency, and node load balancing, to improve the global rationality of routing strategies. The NSGA-II algorithm, combined with an elite retention strategy, ensures both diversity and convergence efficiency in the solution set. The generation of the path priority list takes into account both real-time network status and historical data patterns, making it suitable for dynamic traffic scheduling in large-scale distributed network environments.

[0143] In another technical solution, the process of determining and updating the dynamic threshold in the local routing decision module includes the following steps:

[0144] The local routing decision module loads the baseline threshold T0 when it is initialized. The baseline threshold T0 is calculated based on the normal distribution parameters of historical transmission delay data. The calculation formula is T0 = μ + 3σ, where μ is the mean delay and σ is the standard deviation of the delay.

[0145] The network status detection unit calculates the sliding window average of the link transmission delay between adjacent nodes every five minutes. The window size is the most recent 30 sampling points. When the deviation between the sliding window average and the baseline threshold T0 exceeds 20%, the threshold adaptive adjustment is triggered.

[0146] The threshold adaptive adjustment process performs the following operations:

[0147] Collect the latency data set for each hour within 24 hours before the current time, remove data points with a value exceeding three times the standard deviation, recalculate the mean μ' and standard deviation σ', generate the updated dynamic threshold T1 = μ' + 2.5σ', and overwrite and store it in the local routing decision module;

[0148] If a single delay change is detected between two threshold adjustment cycles and exceeds 150% of the current dynamic threshold, the emergency threshold correction process is immediately initiated, temporarily setting the current dynamic threshold to 80% of the abnormal delay value and marking it as pending calibration.

[0149] When the edge cloud node receives the updated global routing policy issued by the central cloud platform, it synchronously obtains the threshold calibration parameters contained in the global routing policy, and fuses the local dynamic threshold with the calibration parameters through the weighted average algorithm to generate the final dynamic threshold.

[0150] In the above technical solution, the local routing decision module initializes with a baseline threshold of T0 = μ + 3σ (for example, T0 = 80ms when μ = 50ms and σ = 10ms). Historical latency data is analyzed to generate normal distribution parameters by analyzing transmission records from the past seven days. The network status detection unit calculates the sliding window average of link latency between adjacent nodes every five minutes, with the window size consisting of the most recent 30 sampling points (for example, sampling every 30 seconds, with the window covering 15 minutes of data). Adjustments are triggered when the sliding window average deviates from T0 by more than 20% (for example, when the average reaches 96ms).

[0151] For latency data storage, an embedded database supporting the SQLite protocol can be used, while a microcontroller with a floating-point unit can be used for sliding window calculations. Regarding materials, eMMC flash memory can be used for the database storage chip, and FR-4 epoxy resin can be used for the microcontroller substrate. This module should be deployed in the local routing decision module of the edge cloud node and connected to the network interface unit via the SPI bus.

[0152] During the initialization phase, latency data is loaded from the historical database and μ and σ are calculated. T0 is generated and written to non-volatile memory. When the sliding window statistical cycle is triggered, the latest 30 latency sampling points are collected, the arithmetic mean is calculated, and the deviation is compared with T0.

[0153] When the sliding window average deviates from T0 by more than 20%, an adaptive threshold adjustment is triggered. Latency data for the previous 24 hours at the hour is collected (excluding outliers exceeding 3σ), and μ' and σ' are recalculated to generate an updated dynamic threshold, T1 = μ' + 2.5σ' (for example, when μ' = 55ms and σ' = 8ms, T1 = 75ms). If a single delay change exceeds 150% of the current threshold (for example, a delay of 112.5ms is detected when the current threshold T1 = 75ms), an emergency correction is immediately initiated, temporarily setting the threshold to 80% of the outlier value (for example, 90ms), and marking the system as pending calibration.

[0154] Regarding equipment selection, the outlier removal algorithm can be based on a statistical processor that supports Z-score calculations, and the emergency correction module can be equipped with a programmable logic controller (PLC). Regarding materials, the statistical processor's heat sink can be made of aluminum alloy, and the PLC housing can be made of flame-retardant PC plastic. This module needs to be integrated into the threshold management unit of the edge cloud node and interact with the sliding window statistics module via an internal data bus.

[0155] During the adaptive adjustment phase, 24 hours of data is extracted from the latency database, filtered for anomalies, and then the threshold parameters are recalculated. In the event of an emergency correction, a temporary threshold takes effect, overwriting the original value, and a threshold anomaly alert is sent to the central cloud platform. The temporary threshold is reset in the next adjustment cycle or when the global policy is issued.

[0156] When the edge cloud node receives the global routing policy issued by the central cloud platform, it will synchronously obtain the threshold calibration parameters (e.g. calibration coefficient k=0.8) contained in the policy. The local dynamic threshold is fused with the calibration parameters through the weighted average algorithm to generate the final threshold T final =w1×T local + w2×T center (For example, weight w1=0.6, w2=0.4). The weight of the calibration parameters is dynamically allocated according to the success rate of the node's historical strategy execution (for example, when the success rate is ≥90%, w1 is increased to 0.7). final It represents the final dynamic threshold, which is generated by fusing the local threshold with the central platform calibration parameter. local It represents the dynamic threshold of local computing of edge cloud nodes, usually based on statistical analysis of historical delay data. center It represents the global calibration parameters issued by the central cloud platform, which are generated based on the optimization of the whole network status data. w1 and w2 represent the weight coefficients of the local threshold and the central platform parameters, respectively, which are used to balance the impact of the two on the final threshold.

[0157] Regarding device selection, weighted average calculations can be performed using a coprocessor that supports matrix operations, and calibration parameter storage can be stored using an EEPROM chip. Regarding materials, the coprocessor package can be made of a ceramic substrate, and the memory chip pins can be made of gold-plated copper alloy. This module needs to be deployed in the policy execution unit of the edge cloud node and communicate with the central cloud platform via a TLS encrypted link.

[0158] During the working process, the global strategy analysis module extracts the calibration parameters and calculates the weighted average value through the floating point operation unit. The final threshold T final The data is written into the routing table storage unit and synchronized to the threshold management module of the adjacent node. If the calibration parameters conflict with the local threshold, the central platform parameters are prioritized for overwriting.

[0159] This method achieves accurate perception of network latency through baseline threshold initialization and dynamic sliding window statistics. An adaptive adjustment mechanism mitigates threshold failures caused by network environment changes, while an emergency correction function improves the ability to handle sudden anomalies. Global policy synchronization ensures consistency between thresholds and central platform policies, enhancing the coordination and reliability of threshold management in large-scale network environments.

[0160] In another technical solution, the process of periodically collecting network status information includes the following steps:

[0161] The network status detection unit configures multi-dimensional data collection tasks according to the preset collection strategy. The collection strategy includes a 5-minute collection period for link bandwidth utilization, a 30-second collection period for cross-node transmission delay, a 1-minute collection period for network device port status, and a 2-minute collection period for node local computing resource load rate.

[0162] When collecting link bandwidth utilization, ICMP probe packets are sent to adjacent nodes and the arithmetic average of the bandwidth utilization of the last 10 probe results is calculated. The real-time throughput counter value of the network device port is also obtained through the SNMP protocol.

[0163] When performing cross-node transmission delay collection, a UDP benchmark test package is used for bidirectional transmission testing. The average time difference between the sending time and the receiving confirmation time is calculated as the effective delay value, and the maximum delay fluctuation during the test is recorded.

[0164] The collected raw data is processed by the preprocessing module to perform the following operations:

[0165] Abnormal sampling points exceeding three times the standard deviation are eliminated, the binary flag bits in the port status data are converted into one-hot encoding, and indicator data of different dimensions are converted into feature values ​​of uniform magnitude using the Z-score normalization method;

[0166] When the network status detection unit detects that the change rate of two consecutive collection results of any indicator exceeds the preset sensitivity threshold, it automatically shortens the indicator collection cycle to 50% of the original cycle and starts intensive sampling for three consecutive cycles;

[0167] All pre-processed network status information data is attached with a timestamp accurate to the millisecond level when stored, and the clock is synchronized and calibrated with the central cloud platform through the NTP protocol.

[0168] In the above technical solution, the network status detection unit configures a 5-minute collection period for link bandwidth utilization, a 30-second collection period for inter-node transmission delay, a 1-minute collection period for port status, and a 2-minute collection period for node load rate. Link bandwidth utilization is determined by sending ICMP probe packets (e.g., 10 packets per second) to adjacent nodes and calculating the arithmetic mean of the 10 most recent probe results. The unit also obtains the real-time throughput counter value (e.g., a 64-bit counter) of the switch port via SNMP.

[0169] Regarding device selection, the network status detection unit can be equipped with an embedded processor that supports multi-threaded scheduling, and the ICMP detection function can be based on a network interface card that supports the IPv4 / v6 protocols. Regarding materials, the processor substrate can be made of FR-4 epoxy resin, and the interface card housing can be made of aluminum alloy. This module should be deployed in the data acquisition unit of the edge cloud node and connected to the network interface via a PCIe slot.

[0170] During operation, the collection task scheduler triggers each metric collection thread according to a preset period. ICMP probe packets are sent to the target node at regular intervals, and the SNMP protocol retrieves port throughput data using OID identifiers. The collected results are temporarily stored in a ring buffer, awaiting processing by the preprocessing module.

[0171] Cross-node transmission latency is collected using a UDP benchmark test packet for bidirectional transmission testing. The average time difference between the send and receive confirmation times is calculated (e.g., the average of 10 tests), and the maximum latency fluctuation (e.g., a single fluctuation exceeding 50ms) is recorded. The preprocessing module removes outliers from the raw data (sampling points exceeding three standard deviations), converts the port status binary flags to one-hot encoding (e.g., "port enabled" is encoded as [1,0]), and uses Z-score normalization to convert metrics of different dimensions (e.g., bandwidth in Mbps, latency in milliseconds) to a unified scale.

[0172] Regarding equipment selection, UDP testing can be performed on a network card that supports hardware timestamps, and the preprocessing module can run on a coprocessor with a floating-point accelerator. Regarding materials, the network card's PCB can be made of glass fiber reinforced substrate, and the coprocessor's heat sink can be made of copper. This module needs to be integrated into the data processing unit of the edge cloud node and interact directly with the acquisition unit via a DMA channel.

[0173] During operation, UDP test packets carry nanosecond timestamps. The receiver calculates round-trip latency and filters out outliers caused by network jitter (for example, data points with latency > 200ms). The preprocessing module encodes the port status and normalizes it, mapping bandwidth utilization to the [-1, 1] range and latency to the [0, 1] range, facilitating unified processing by subsequent analysis modules.

[0174] If the rate of change between two consecutive collection results for any metric exceeds a preset sensitivity threshold (e.g., bandwidth utilization change ≥ 15%), the metric collection cycle is automatically shortened to 50% of the original cycle (e.g., from 5 minutes to 2.5 minutes), and three consecutive cycles of intensive sampling are initiated. All preprocessed data is timestamped with millisecond accuracy (e.g., 2023-10-01T12:34:56.789) and synchronized with the central cloud platform via the NTP protocol (error ≤ 1ms).

[0175] In terms of device selection, clock synchronization can be based on a hardware clock module that supports the PTP protocol, and dynamic period adjustment can be implemented using a programmable timer chip. Regarding materials, the clock module's crystal oscillator can be made of quartz, and the timer pins can be made of gold-plated copper alloy. This module should be deployed in the clock management unit of the edge cloud node and connected to the data storage unit via the I2C bus.

[0176] During operation, the rate-of-change detection module calculates the difference between indicators in real time (for example, (current value - previous value) / previous value). If the difference exceeds the threshold, a period adjustment instruction is triggered. During intensive sampling, the collection frequency is increased to 50% of the original period and restored to the default setting after three periods. The timestamp generator synchronizes time with an NTP server to ensure time consistency across node data.

[0177] This method improves the real-time and accuracy of network status data through multi-dimensional periodic collection and dynamic adjustment. Standardization and anomaly filtering reduce noise interference and enhance the reliability of subsequent analysis. Clock synchronization and timestamp calibration ensure the temporal consistency of data across nodes, making it suitable for time-sensitive, large-scale distributed network monitoring scenarios.

[0178] In another technical solution, the processing after the suboptimal candidate path verification fails further includes the following steps:

[0179] If the candidate path fails verification three times in a row, the local routing decision module performs the following operations:

[0180] Roll back the traffic on the current faulty link to the path before the switchover and generate a path rollback instruction to update the forwarding table entry.

[0181] Send a path switching failure alarm to the central cloud platform. The alarm includes the failed path identifier, verification failure reason code, and current node load status.

[0182] Activate local degradation mode to limit the forwarding rate of non-critical business traffic to a preset safety threshold while maintaining the minimum guaranteed bandwidth for critical business traffic;

[0183] After receiving the path switching failure alarm information, the central cloud platform marks the path as a high-risk path in the updated global routing policy and excludes it from the failover path set for at least 24 hours.

[0184] In the above technical solution, if candidate path verification fails three times in a row (for example, packet loss rate ≥ 3% or latency difference ≥ 15ms), the local routing decision module will roll back traffic on the failed link to the pre-switching path and generate a path rollback instruction to update the forwarding table entry. The alarm information includes the failed path identifier (for example, path ID = 0x5A), the verification failure reason code (for example, code 0x01 indicates latency exceeded), and the current node load status (for example, CPU load rate 75%). The alarm information is uploaded to the central cloud platform via a TLS encrypted tunnel, with an upload frequency of once per second until a confirmation response is received from the platform.

[0185] In terms of device selection, path rollback operations can be performed using an SDN controller that supports the OpenFlow protocol, and alarm information encapsulation can be implemented using an embedded processor that supports JSON serialization. Regarding materials, the controller circuit board can be constructed using FR-4 epoxy resin, and the processor package can be constructed using ceramic. This module should be deployed in the local routing decision module of the edge cloud node and connected to the routing table storage unit via a high-speed bus.

[0186] During operation, the rollback command triggers the restoration of the forwarding table entry to the historical version, and the changes are synchronized to adjacent nodes via the BGP protocol. After the alarm information is generated, a timestamp and digital signature are added to ensure data integrity and source authenticity.

[0187] When local degradation mode is enabled, the forwarding rate of non-critical service traffic (such as video streaming) is limited to a preset safety threshold (for example, 10 Mbps), while critical service traffic (such as VoIP) maintains a minimum guaranteed bandwidth (for example, 2 Mbps). Rate limiting is implemented through QoS policies, such as using a token bucket algorithm to control traffic shaping, with a token generation rate of 1 Mbps / ms.

[0188] In terms of device selection, traffic control can be based on switch chips that support the DiffServ protocol, and the token bucket algorithm can be implemented on a network processor equipped with a hardware counter. Regarding materials, aluminum alloy can be used for the switch chip's heat sink, and gold-plated copper alloy can be used for the processor pins. This module needs to be integrated into the traffic management unit of the edge cloud node and communicate with the routing decision module via a PCIe interface.

[0189] During operation, the traffic classification module distinguishes traffic types based on the DSCP field. Non-critical traffic is marked as low priority and placed in the rate-limited queue. Critical traffic is directly placed in the high-priority queue, ensuring that minimum bandwidth is not degraded. Rate-limiting parameters are dynamically configured through the CLI or SNMP.

[0190] After receiving the alert, the central cloud platform marks the failed path as a high-risk path (for example, red) in the updated global routing policy and excludes it from the set of failover paths for at least 24 hours. During this exclusion period, the path is not included in the candidate path prioritization until it is manually unmarked or automatically expires.

[0191] Regarding equipment selection, high-risk path management can be based on a network topology management system that supports a graph database, and path exclusion policies can be implemented using a rules engine module. Regarding materials, solid-state drives (SSDs) can be used as database storage media, and copper heat pipes can be used as the heat sink for the rules engine. This module should be deployed in the policy management node of the central cloud platform and connected to the global routing analysis engine via Gigabit Ethernet.

[0192] During operation, the policy management node parses the path ID in the alarm information and updates the path status attributes in the graph database. Dynamic updates to the failover path set are performed every five minutes to ensure that high-risk paths are promptly excluded. After the exclusion period expires, the path automatically rejoins the candidate pool and participates in subsequent verification.

[0193] This method uses path fallback and alert mechanisms to minimize the impact of service interruptions and ensure the continuity of core services. Local degradation mode prioritizes critical services when resources are limited, enhancing the tiered assurance capabilities of network services. Dynamically eliminating high-risk paths reduces the risk of repeated handover failures and enhances the robustness of the fault recovery process, making it suitable for edge network scenarios requiring high reliability.

[0194] The present invention provides a distributed large-scale Layer 2 network intelligent routing system based on edge cloud nodes, comprising:

[0195] Multiple edge cloud nodes and central cloud platforms. The edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol;

[0196] Each edge cloud node includes a network status detection unit, a local routing decision module, and a routing table storage unit. The network status detection unit periodically collects link bandwidth utilization, cross-node transmission delay, network device port status, and node local computing resource load rate between adjacent nodes;

[0197] The local routing decision module encapsulates the data collected by the network status detection unit into structured data packets and transmits them to the central cloud platform through an encrypted tunnel;

[0198] The central cloud platform is equipped with a global routing analysis engine. Based on the network status information received from all edge cloud nodes, the global routing analysis engine uses a spatiotemporal feature extraction neural network model to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths.

[0199] After receiving the global routing policy issued by the central cloud platform, the local routing decision module updates the forwarding entries in the routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding entries to the adjacent edge cloud nodes through the pre-configured communication protocol;

[0200] The network status detection unit monitors the variation of link transmission delay between adjacent nodes in real time. When it detects that the variation exceeds the dynamic threshold pre-stored in the routing table storage unit, it triggers the local routing decision module to reconfigure the routing table storage unit based on the failover path set and upload the marked link status anomaly information to the central cloud platform via an encrypted tunnel.

[0201] The global routing analysis engine of the central cloud platform initiates policy recalculation based on the link status anomaly information and distributes the updated global routing policy to the associated affected edge cloud nodes.

[0202] In the above technical solution, edge cloud nodes establish a bidirectional connection with the central cloud platform via an IPsec VPN tunnel. The communication protocol can support MQTT or gRPC, and the heartbeat detection period can be set to 10 seconds. Each edge cloud node includes a network status detection unit (which periodically collects link bandwidth utilization, latency, etc.), a local routing decision module (which performs path switching decisions), and a routing table storage unit (which stores forwarding table entries).

[0203] In terms of equipment selection, edge cloud nodes can utilize industrial-grade servers equipped with dual network ports, and the network status detection unit can be equipped with an embedded controller supporting SNMP and ICMP protocols. Regarding materials, the server chassis can be constructed of aluminum alloy, with internal circuit boards constructed of FR-4 epoxy resin. Edge cloud nodes are deployed at the edge of the network, close to end devices, such as in a smart factory workshop switch cabinet, and are directly connected to adjacent nodes via Cat6a network cables.

[0204] During operation, edge cloud nodes automatically establish an encrypted tunnel with the central cloud platform upon startup and periodically send heartbeat packets to maintain the connection. The network status detection unit collects data at a preset interval (for example, bandwidth is collected every 5 minutes). The local routing decision module receives the policy issued by the central platform, updates the routing table, and synchronizes the changes to neighboring nodes via the BGP protocol.

[0205] The central cloud platform's global routing analysis engine uses a parallel model of a convolutional neural network (CNN) and a long short-term memory network (LSTM). Its input includes node topology and delay sequences, and its output is a list of priority paths (for example, priority 1 is path A, with a weight of 70%). After the policy is generated, it is encapsulated in JSON format and sent to edge cloud nodes over a TLS encrypted link, with delivery latency kept to under 200ms.

[0206] In terms of equipment selection, the global routing analysis engine can run on a server cluster supporting GPU acceleration, and the neural network model can be implemented based on the TensorFlow framework. Regarding materials, the server cooling module can use a combination of copper heat pipes and aluminum alloy fins, and the data storage unit can be equipped with an NVMe solid-state drive. The central cloud platform is deployed in the data center computer room and connected to the edge nodes via redundant fiber optic links.

[0207] During operation, the central cloud platform receives status data from all edge nodes. The neural network model extracts spatiotemporal features and generates a routing policy. The policy file is compressed, signed, and distributed to the target node. The local routing decision module of the edge node parses the JSON instructions and updates the forwarding table, while simultaneously sending synchronization requests to adjacent nodes.

[0208] Edge cloud nodes monitor the transmission delay variation between adjacent links in real time, with a dynamic threshold initialized to T0 = μ + 3σ (for example, T0 = 80ms when μ = 50ms and σ = 10ms). When the delay variation exceeds the threshold (for example, reaching 100ms), the local routing decision module is triggered to select a candidate path (for example, path B) from the failover path set. The module then performs packet loss rate detection (with a 2% threshold) and latency verification (with a ±10ms tolerance). If verification fails, an exception message (including the delay value, path ID, and node load status) is sent to the central cloud platform.

[0209] In terms of equipment selection, latency monitoring can be based on a network card that supports hardware timestamping, while failover verification can be performed using a probe module that supports the IP SLA protocol. Regarding materials, the network card's PCB substrate can be made of fiberglass, and the probe housing can be made of flame-retardant ABS plastic. This module needs to be integrated into the network interface unit of the edge cloud node and communicate with the routing decision module via the PCIe interface.

[0210] During operation, link status anomalies trigger local path switching. After candidate paths pass verification, traffic is diverted (for example, 70% of traffic is switched to path B). If verification fails three times in a row, the system falls back to the original path and limits bandwidth for non-critical services to 10 Mbps. Exception information is timestamped and uploaded via an encrypted tunnel. Upon receipt, the central platform initiates global policy recalculation.

[0211] This system utilizes a distributed architecture and a central coordination mechanism to achieve real-time perception and dynamic optimization of network status. Local decision-making at edge nodes reduces computing pressure on the central platform, while failover processes ensure business continuity. Encrypted communications and policy signatures enhance data transmission security, making it suitable for the high-reliability network operations and maintenance requirements of large-scale edge computing scenarios.

[0212] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A distributed large-scale layer 2 network intelligent routing method based on edge cloud nodes, characterized in that: The following steps are involved: Multiple edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol. Each edge cloud node periodically collects network status information of the node where it is located. The network status information includes link bandwidth utilization between adjacent nodes, cross-node transmission delay, network device port status, and node local computing resource load rate; Each edge cloud node encapsulates the network status information into a structured data packet and uploads it to the central cloud platform via an encrypted tunnel; Based on the network status information of all edge cloud nodes, the central cloud platform uses a neural network model based on spatiotemporal feature extraction to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths between edge cloud nodes. After receiving the global routing policy issued by the central cloud platform, each edge cloud node updates the forwarding table entries in the local routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding table entries to the adjacent nodes through the pre-configured communication protocol; Each edge cloud node monitors the link transmission delay variation between adjacent nodes in real time. When it detects that the link transmission delay variation exceeds the dynamic threshold pre-stored in the local routing decision module, it triggers the local routing decision module to reconfigure the local routing table based on the failover path set in the global routing policy, and marks the link status anomaly information and uploads it to the central cloud platform via an encrypted tunnel. The central cloud platform initiates global routing policy recalculation based on the link status anomaly information and distributes the updated global routing policy to the affected edge cloud nodes associated with the link status anomaly information; The neural network model based on spatiotemporal feature extraction includes a convolutional neural network branch and a long short-term memory network branch connected in parallel, wherein: The convolutional neural network branch uses the physical topological connection relationship of edge cloud nodes as input and extracts spatial correlation features between adjacent nodes through three convolutional layers. The spatial correlation features include the link bandwidth fluctuation pattern and port status change trend between nodes. The long short-term memory network branch takes the cross-node transmission delay sequence as input and extracts the delay fluctuation time series characteristics within a preset time window through a bidirectional loop structure. The delay fluctuation time series characteristics include periodic congestion patterns and sudden abnormal delay segments; The neural network model dynamically assigns weights to spatial correlation features and delay fluctuation time series features through the attention mechanism layer to generate a network state vector that integrates spatiotemporal features. The fully connected layer of the neural network model generates a global routing strategy based on the network state vector. The generation process of the path priority list involves performing a multi-objective optimization operation on the network state vector that integrates spatiotemporal features. The constraints of the multi-objective optimization operation include the link bandwidth utilization threshold, the transmission delay upper limit, and the node load balancing coefficient. The process of determining and updating the dynamic threshold in the local routing decision module includes the following steps: The local routing decision module loads the baseline threshold T0 when it is initialized. The baseline threshold T0 is calculated based on the normal distribution parameters of historical transmission delay data. The calculation formula is T0 = μ + 3σ, where μ is the mean delay and σ is the standard deviation of the delay. The network status detection unit calculates the sliding window average of the link transmission delay between adjacent nodes every five minutes. The window size is the most recent 30 sampling points. When the deviation between the sliding window average and the baseline threshold T0 exceeds 20%, the threshold adaptive adjustment is triggered. The threshold adaptive adjustment process performs the following operations: Collect the latency data set for each hour within 24 hours before the current time, remove data points with a value exceeding three times the standard deviation, recalculate the mean μ' and standard deviation σ', generate the updated dynamic threshold T1 = μ' + 2.5σ', and overwrite and store it in the local routing decision module; If a single delay change is detected to exceed 150% of the current dynamic threshold between two threshold adjustment cycles, the emergency threshold correction process is immediately initiated, temporarily setting the current dynamic threshold to 80% of the abnormal delay value and marking it as pending calibration. When the edge cloud node receives the updated global routing policy issued by the central cloud platform, it synchronously obtains the threshold calibration parameters contained in the global routing policy, and fuses the local dynamic threshold with the calibration parameters through the weighted average algorithm to generate the final dynamic threshold.

2. The distributed large-scale layer 2 network intelligent routing method based on edge cloud nodes according to claim 1 is characterized in that: Reconfiguring the local routing table storage unit based on the failover path set includes the following steps: When it is detected that the link transmission delay variation exceeds the dynamic threshold, a pre-sorted candidate path sequence is extracted from the failover path set, and the candidate path sequence is prioritized based on the historical transmission success rate and the number of path hops; The highest-priority candidate path is tested for link bandwidth availability and transmission delay. Link bandwidth availability is tested by sending probe packets to the target node and calculating the packet loss rate. Transmission delay is verified by measuring the difference between the round-trip delay of the probe packets and a preset delay threshold. If the packet loss rate of the candidate path is lower than the preset packet loss threshold and the round-trip delay difference is within the preset tolerance range, the traffic of the current faulty link is diverted to the candidate path according to the preset ratio, and a forwarding table entry update instruction including the diversion ratio and path identifier is generated; If the candidate path verification fails, the dynamic weight adjustment of the candidate path sequence is triggered, the path priority is recalculated according to the real-time load rate of the adjacent nodes, and the link bandwidth availability detection step is jumped to perform the suboptimal candidate path verification; After the path switch is completed, the local routing decision module stores the updated forwarding table entry in association with the failover timestamp and sends a routing table synchronization request to the adjacent edge cloud node through a preconfigured communication protocol. The synchronization request includes the changed forwarding table entry hash value and version number. The network status detection unit continuously monitors the transmission quality indicators of the switched path. When the transmission delay fluctuation exceeds 50% of the dynamic threshold within three consecutive detection cycles, the secondary path switching process is triggered and the priority weight parameters of the failover path set are updated.

3. The distributed large layer 2 network intelligent routing method based on edge cloud nodes according to claim 1 is characterized in that: The execution process of the multi-objective optimization operation includes the following steps: The network state vector that integrates spatiotemporal features is mapped to a bandwidth allocation scaling factor α i , delay compensation factor β j and node load balancing weight γ k A set of decision variables, where i corresponds to the path number, j corresponds to the delay level identifier, and k corresponds to the node identifier; Construct an objective function set including the first objective function, the second objective function and the third objective function. The first objective function is f1=Σ(α i ×C i ), used to maximize the effective bandwidth utilization, the second objective function is f2=Σ(β j ×D j ), used to minimize the transmission delay offset, the third objective function is f3=Σ(γ k ×L k ), used to balance the node load differences, where C i represents the available bandwidth capacity of path i, D j represents the delay compensation amount of path j, L k represents the load deviation value of node k; Set up a system of equations with the following constraints: Σα i ≤ predefined link bandwidth utilization threshold, Σβ j ≤predefined transmission delay upper limit, Σγ k ≤predefined node load balancing coefficient; A non-dominated sorting genetic algorithm with an elite retention strategy is used to solve the objective function set, which includes: An initial solution set is generated based on the path priority list as the genetic algorithm population. An adaptive crossover operation based on path similarity is performed. The crossover probability is dynamically adjusted according to the path overlap rate of the individuals in the solution set. An adjacent path replacement mutation operation is performed to replace the current path with a candidate path that is directly adjacent in the physical topology. The following processing is performed on the solution set produced by each iteration: Delete the individuals in the equation group that violate the constraints, perform non-dominated sorting on the remaining individuals and calculate the crowding distance, and retain the first preset number of individuals with the highest sorting level and the largest crowding distance; From the Pareto frontier solution set of the final iterative result, select the individual with the largest crowding distance as the optimal solution, and extract the α corresponding to the individual. i , β j and γ k The parameters are combined to generate a ranking weight value of the path priority list.

4. The distributed large-scale layer 2 network intelligent routing method based on edge cloud nodes according to claim 1 is characterized in that: The process of periodically collecting network status information includes the following steps: The network status detection unit configures multi-dimensional data collection tasks according to the preset collection strategy. The collection strategy includes a 5-minute collection period for link bandwidth utilization, a 30-second collection period for cross-node transmission delay, a 1-minute collection period for network device port status, and a 2-minute collection period for node local computing resource load rate. When collecting link bandwidth utilization, ICMP probe packets are sent to adjacent nodes and the arithmetic average of the bandwidth utilization of the last 10 probe results is calculated. The real-time throughput counter value of the network device port is also obtained through the SNMP protocol. When performing cross-node transmission delay collection, a UDP benchmark test package is used for bidirectional transmission testing. The average time difference between the sending time and the receiving confirmation time is calculated as the effective delay value, and the maximum delay fluctuation during the test is recorded. The collected raw data is processed by the preprocessing module to perform the following operations: Abnormal sampling points exceeding three times the standard deviation are eliminated, the binary flag bits in the port status data are converted into one-hot encoding, and indicator data of different dimensions are converted into feature values ​​of uniform magnitude using the Z-score normalization method; When the network status detection unit detects that the change rate of two consecutive collection results of any indicator exceeds the preset sensitivity threshold, it automatically shortens the indicator collection cycle to 50% of the original cycle and starts intensive sampling for three consecutive cycles; All pre-processed network status information data is attached with a timestamp accurate to the millisecond level when stored, and the clock is synchronized and calibrated with the central cloud platform through the NTP protocol.

5. The distributed large-scale layer 2 network intelligent routing method based on edge cloud nodes according to claim 2 is characterized in that: The processing after the suboptimal candidate path verification fails also includes the following steps: If the candidate path fails verification three times in a row, the local routing decision module performs the following operations: Roll back the traffic on the current faulty link to the path before the switchover and generate a path rollback instruction to update the forwarding table entry. Send a path switching failure alarm to the central cloud platform. The alarm includes the failed path identifier, verification failure reason code, and current node load status. Activate local degradation mode to limit the forwarding rate of non-critical business traffic to a preset safety threshold while maintaining the minimum guaranteed bandwidth for critical business traffic; After receiving the path switching failure alarm information, the central cloud platform marks the path as a high-risk path in the updated global routing policy and excludes it from the failover path set for at least 24 hours.

6. A distributed large-scale layer 2 network intelligent routing system based on edge cloud nodes, characterized in that: The method for executing the distributed large-scale Layer 2 network intelligent routing method based on edge cloud nodes according to any one of claims 1 to 5 comprises: Multiple edge cloud nodes and central cloud platforms. The edge cloud nodes establish a two-way data connection with the central cloud platform through a pre-configured communication protocol; Each edge cloud node includes a network status detection unit, a local routing decision module, and a routing table storage unit. The network status detection unit periodically collects link bandwidth utilization, cross-node transmission delay, network device port status, and node local computing resource load rate between adjacent nodes; The local routing decision module encapsulates the data collected by the network status detection unit into structured data packets and transmits them to the central cloud platform through an encrypted tunnel; The central cloud platform is equipped with a global routing analysis engine. Based on the network status information received from all edge cloud nodes, the global routing analysis engine uses a spatiotemporal feature extraction neural network model to generate a global routing strategy. The global routing strategy includes a path priority list, traffic distribution weight coefficients, and a set of failover paths. After receiving the global routing policy issued by the central cloud platform, the local routing decision module updates the forwarding entries in the routing table storage unit according to the path priority list and traffic distribution weight coefficient, and synchronizes the updated forwarding entries to the adjacent edge cloud nodes through the pre-configured communication protocol; The network status detection unit monitors the variation of link transmission delay between adjacent nodes in real time. When it detects that the variation exceeds the dynamic threshold pre-stored in the routing table storage unit, it triggers the local routing decision module to reconfigure the routing table storage unit based on the failover path set and upload the marked link status anomaly information to the central cloud platform via an encrypted tunnel. The global routing analysis engine of the central cloud platform initiates policy recalculation based on the link status anomaly information and distributes the updated global routing policy to the associated affected edge cloud nodes.

Images