A network congestion mitigation method based on network resource adjustment

Abstract

This invention discloses a network congestion mitigation method based on network resource adjustment. First, it identifies traffic patterns and quantifies congestion indicators by analyzing historical and real-time data. Then, it uses an LSTM model to predict the probability of future traffic surges. When the probability exceeds a threshold, relay nodes are automatically grouped by capability, and an optimized computing power allocation scheme is generated. Subsequently, network paths are dynamically adjusted to redirect traffic, and path stability is evaluated. If stability is not met, iterative optimization is initiated, updating the allocation scheme to ensure reliability. Finally, a stable and reliable scheduling configuration is deployed to the production environment, and the effect is verified through continuous monitoring. Feedback data is then injected into the next round of analysis, forming a self-optimizing closed-loop management system.

Description

Technical Field

[0001] This invention relates to the field of dynamic resource adjustment, and more particularly to a network congestion mitigation method for adjusting network resources. Background Technology

[0002] In modern communication networks, subnet scheduling, as a key technology for optimizing resource allocation and improving network performance, plays a crucial role in ensuring data transmission efficiency and stability. With the rapid growth of user demands and the diversification of network applications, hotspot areas within the subnet—where traffic is concentrated due to high-frequency access to certain nodes or regions—have become a core issue affecting overall network performance.

[0003] Especially in high-load scenarios, such as large events or sudden incidents, traffic surges in hotspot areas can lead to congestion, resulting in delays or even service interruptions. Therefore, efficiently managing hotspot areas within a subnet and ensuring dynamic resource allocation has become a major challenge that urgently needs to be addressed in the field of network scheduling. Existing methods for dealing with traffic congestion in hotspot areas often rely on static resource allocation strategies or simple real-time monitoring. These methods are typically based on fixed thresholds or rules for scheduling, making it difficult to adapt to rapid changes in traffic patterns.

[0004] For example, when traffic in hotspot areas suddenly increases due to unforeseen events, existing systems may fail to identify potential risks in a timely manner and lack the ability to flexibly adjust resource allocation. Furthermore, existing methods often ignore the dynamic characteristics of time periods and user behavior changes when dealing with complex network environments, leading to inefficient resource allocation and frequent congestion in hotspot areas. In subnet scheduling, the core technical challenge lies in accurately predicting traffic changes in hotspot areas and dynamically optimizing resource allocation. The first key factor is the highly dynamic and uncertain nature of access patterns in hotspot areas. Access patterns are not only affected by time periods (such as traffic differences between weekdays and weekends) but also fluctuate drastically due to unforeseen events or changes in user behavior.

[0005] For example, during a large-scale live broadcast, traffic in a specific area may surge in a short period, and existing scheduling systems struggle to quickly identify such pattern changes. The second key factor is the insufficient allocation of processing capacity to remaining relay nodes. Backup relay nodes in the network may be in standby mode, but when traffic surges in hotspot areas, quickly assessing the availability and allocation efficiency of these nodes becomes a bottleneck in resource scheduling. These two factors are interconnected: the dynamic nature of access patterns requires the system to have rapid predictive capabilities, and the accuracy of these predictions directly affects the allocation efficiency of relay nodes. If traffic changes cannot be predicted in a timely manner, the activation of backup nodes and path optimization will lag, further exacerbating congestion. Therefore, how to quickly assess and allocate the processing capacity of remaining relay nodes based on the dynamic access patterns of hotspot areas in subnet scheduling becomes a key issue for network performance optimization. Summary of the Invention

[0006] The purpose of this invention is to provide a network congestion mitigation method for adjusting network resources, which solves the aforementioned technical problems pointed out in the prior art.

[0007] This invention provides a network congestion mitigation method for adjusting network resources, comprising the following steps:

[0008] By collecting historical traffic data and current access logs of each node in the subnet in real time, and using time series analysis methods, dynamic pattern characteristics of hotspot areas are extracted from historical traffic data and current access logs to obtain preliminary traffic congestion indicators.

[0009] Based on preliminary traffic congestion indicators, a long short-term memory network is applied to model access prediction for hotspot areas, resulting in a traffic prediction model; the probability of future traffic surges is determined using the traffic prediction model.

[0010] If the probability of a future traffic surge exceeds a preset threshold, the processing capacity information of the remaining relay nodes is obtained from the network topology. Based on the processing capacity information of the relay nodes and the probability of a future traffic surge, the remaining relay nodes are grouped using the K-means clustering algorithm to obtain multiple node grouping results.

[0011] The available bandwidth and latency parameters of relay nodes in the grouping results of each node are collected. Based on the available bandwidth and latency parameters of relay nodes and the probability of future traffic surges, traffic information is allocated through particle swarm optimization combined with node state traffic impact analysis to obtain an optimized computing power allocation scheme.

[0012] Based on the optimized computing power allocation scheme, the subnet scheduling path is dynamically adjusted, and the access prediction results are integrated to reroute the traffic data flow to the allocated relay node, and the stability score of the adjusted path is determined.

[0013] If the adjusted path stability score is determined to be lower than the preset stability threshold, the available bandwidth of the remaining relay nodes will be reassessed through an iterative optimization process, and the allocation scheme will be updated to minimize the risk of traffic congestion, resulting in the final scheduling configuration.

[0014] Based on the final scheduling configuration, it is deployed to the subnet to perform resource allocation and path rerouting operations in real time, and to perform real-time monitoring and processing operations.

[0015] Preferably, based on the available bandwidth and latency parameters of relay nodes and the probability of future traffic surges, a traffic information allocation scheme is obtained by combining particle swarm optimization with node state traffic impact analysis, including the following steps:

[0016] Initialize particle swarm parameters, including determining hotspot nodes based on the probability of future traffic surges; establishing a relay node set based on all relay nodes; establishing a hotspot node set based on all hotspot nodes; and dividing the network topology into multiple partitions based on the location of each node, with each partition containing multiple hotspot nodes and multiple relay nodes.

[0017] Traverse each partition and perform the following processing: Establish an allocation matrix A based on the relay node set G and the hotspot node set H; each element aij in the allocation matrix represents the allocation of the i-th relay node... The computing resources are allocated to the j-th hotspot node. Multiple allocation result individuals are obtained based on each element aij in the allocation matrix A;

[0018] The initial fitness F is calculated based on the probability of future traffic surges for each hotspot node in each assigned individual, the available bandwidth and latency parameters of each relay node, and the network topology distance between the hotspot node and the relay node.

[0019] Multiple candidate computing power allocation schemes are obtained by screening based on the initial optimization fitness F; the traffic surge caused by the probability of future traffic surges is analyzed for the candidate computing power allocation schemes, and the traffic impact on hot nodes and relay nodes in the candidate computing power allocation schemes is analyzed to obtain the load transition amplitude.

[0020] An optimized computing power allocation scheme is obtained by iteratively filtering the load transition amplitude using a preset first load transition amplitude threshold and a second load transition amplitude threshold.

[0021] The preferred delay parameters for relay node i This represents the processing delay value for the relay node.

[0022] Preferably, the candidate computing power allocation schemes are selected based on the initial optimization fitness F, and the top n computing power allocation schemes in terms of fitness are retained as candidate computing power allocation schemes.

[0023] Preferably, an optimized computing power allocation scheme is obtained by iteratively judging and filtering the load transition amplitude using a preset first load transition amplitude threshold and a second load transition amplitude threshold, including the following operation steps:

[0024] When the load transition amplitude corresponding to the current candidate computing power allocation scheme is greater than or equal to the first load transition amplitude threshold, the candidate computing power allocation schemes corresponding to the load transition amplitude being greater than or equal to the first load transition amplitude threshold and less than the second load transition amplitude threshold are updated according to the preset particle search step size to obtain a new candidate computing power allocation scheme. The new candidate computing power allocation scheme is then returned to the above allocation result individual and iterated again until the optimized computing power allocation scheme is output.

[0025] Candidate computing power allocation schemes whose load transition amplitude is greater than the second load transition amplitude threshold are filtered out.

[0026] When the load transition amplitude corresponding to the current candidate computing power allocation scheme is less than the first load transition amplitude threshold, the corresponding allocation scheme to be determined is output as the optimized computing power allocation scheme.

[0027] Preferably, the candidate computing power allocation scheme is analyzed in conjunction with the probability analysis of future traffic surges. The resulting traffic surges will impact the hotspot nodes and relay nodes in the candidate computing power allocation scheme, and the load transition amplitude is obtained. This includes the following steps:

[0028] The traffic surge time points are obtained based on the future traffic surge probability of each hotspot node; the expected traffic corresponding to the future traffic surge probability of each hotspot node is obtained; the allocated carrying traffic of each relay node at each traffic surge time point is determined based on the node grouping results of each relay node corresponding to each hotspot node in each candidate computing power allocation scheme; the load rate of the relay node at the traffic surge time point is obtained by analyzing the allocated carrying traffic and the available bandwidth and latency parameters of the relay node.

[0029] Calculate the load change magnitude at traffic surge time point t and traffic surge time point t+1 for the i-th relay node; determine the load change rate of the relay node at traffic surge time point t based on the load change magnitude;

[0030] Obtain the topological distance attenuation factor and traffic interaction intensity between hot nodes and the i-th relay node in the current candidate computing power allocation scheme;

[0031] The basic impact contribution is calculated based on the load change rate, flow interaction intensity, and topological distance attenuation factor.

[0032] The load transition amplitude is calculated based on the basic impact contribution and the expected flow amplitude corresponding to the expected flow.

[0033] A preferred method for calculating the load transition amplitude is to calculate the absolute value of the change in the basic impact contribution of all relay nodes and hotspot nodes at consecutive time points, and then sum them up by weighting the expected flow amplitude and the time weight factor to obtain the load transition amplitude.

[0034] Ideally, the allocation scheme is updated to minimize the risk of traffic congestion, resulting in the final scheduling configuration, including the allocation of computing resources for relay nodes and hotspot nodes.

[0035] Preferably, the allocation of computing resources between relay nodes and hotspot nodes to obtain the final scheduling configuration includes the following steps:

[0036] Iterate through each relay node corresponding to each hotspot node, and set priority weights for each relay node according to the business processing type of the relay node.

[0037] The computing power allocation ratio of each relay node to the corresponding hot spot node is calculated based on the available bandwidth, priority weight, and adjusted path stability score of each relay node.

[0038] The server obtains a list of hot nodes as a whole, and calculates the priority scheduling evaluation value of each hot node based on the probability of future traffic surges, the business importance level of each hot node, and the real-time congestion level of the server.

[0039] The final scheduling configuration is determined based on the computing power allocation ratio of each relay node to the corresponding hot spot node, the priority scheduling evaluation value of each hot spot node, and the expected traffic corresponding to the probability of future traffic surges.

[0040] In another aspect, the present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the network congestion mitigation method for adjusting network resources described above.

[0041] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0042] Analysis of the network congestion mitigation method for adjusting network resources provided by this invention reveals that, in practical applications, it first collects two types of time-series data: historical traffic data and current access logs. Time-series analysis methods are then used to mine patterns from the time dimension, identifying hotspot areas (i.e., network areas with abnormal or persistently high traffic, such as application server clusters or database clusters experiencing abnormal traffic surges). Dynamic pattern characteristics reflecting these traffic changes (such as periodic peaks or trending increases) are then extracted. Finally, these characteristics are quantified into preliminary traffic congestion indicators, transforming the raw, chaotic, massive network data... The data is transformed into structured indicators that can quantify and assess network status. Furthermore, by receiving preliminary traffic congestion indicators and using a Long Short-Term Memory (LSTM) network for modeling, the LSTM not only learns short-term dependencies but also captures long-term patterns. By fusing time-period (e.g., weekdays / weekends) and user behavior data, the model can simulate the potential impact of scenarios such as promotions and unexpected events on traffic, outputting the probability of future traffic surges. This achieves a leap from current situation perception to future prediction, proactively anticipating congestion risks. Furthermore, by setting a preset threshold as an action trigger, if the predicted probability of a traffic surge exceeds this threshold... If necessary, intervention is required. The processing capacity information of all remaining relay nodes is retrieved from the network topology. K-means clustering is then used to group nodes in the subnet according to their processing capacity, simplifying the large number of nodes into several resource pools with different processing capacity levels, preparing for precise matching in the next step. Furthermore, based on the node grouping results, key performance indicators such as available bandwidth and latency parameters of nodes within each group are calculated. Based on the available bandwidth and latency parameters, an optimized computing power allocation scheme is generated. Through a priority matching principle, the most scarce resources (high-processing-capacity nodes) are ensured to be used where they are most needed. (High congestion area) thereby theoretically maximizing resource utilization efficiency and achieving effective load balancing; further, based on the optimized computing power allocation scheme, the subnet scheduling path is dynamically adjusted physically. Specifically, the access prediction results are integrated to reroute traffic data flow to the allocated relay node, actively determine the stability score of the adjusted path, assess whether this path change introduces new risks (such as single point of failure, excessively long path, etc.), complete the actual redirection of traffic, transform the prediction and allocation scheme into actual network state changes, and avoid secondary problems of oscillation or instability caused by solving congestion through path stability.

[0043] Furthermore, when the assessment finds that the adjusted path stability score is lower than the preset stability threshold, it indicates that the initial allocation scheme is not ideal. In this case, an iterative optimization process is initiated to reassess the processing capacity of the nodes (which may be due to changes in network status) and update the allocation scheme, forming an internal closed-loop optimization mechanism. This ensures that the final output scheme not only theoretically alleviates congestion but is also stable and reliable in practice. Finally, the determined final scheduling configuration is deployed to the subnet for real-time execution, and real-time monitoring and processing operations are continuously performed. This involves monitoring the network performance indicators after execution and comparing them with the initial target to verify whether traffic congestion in hotspot areas has been alleviated. The results of the entire analysis, prediction, and optimization process are truly applied to the production environment, and the monitoring results (new network performance data) are fed back to the initial steps as part of the next round of historical traffic data. Attached Figure Description

[0044] Figure 1 A schematic diagram of the main process of a network congestion mitigation method for adjusting network resources;

[0045] Figure 2 A schematic diagram simulating node grouping results in a network congestion mitigation method for adjusting network resources;

[0046] Figure 3 A schematic diagram simulating the operation process of allocating an optimized computing power allocation scheme in a network congestion mitigation method for adjusting network resources;

[0047] Figure 4 This is a schematic diagram simulating the computing power scheduling configuration in a network congestion mitigation method that adjusts network resources. Detailed Implementation

[0048] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0050] Example 1

[0051] like Figure 1 As shown, Embodiment 1 of the present invention provides a network congestion mitigation method for adjusting network resources, including the following steps:

[0052] Step S10: By collecting historical traffic data and current access logs from each node in the subnet in real time, time series analysis is used to extract dynamic pattern characteristics of hotspot areas from the historical traffic data and current access logs to obtain preliminary traffic congestion indicators. The so-called hotspot network area can be an application server cluster area or a database cluster: Application server cluster: This area handles core business logic, such as order placement and payment. Database cluster: This area stores all product, user, and order data. The aforementioned hotspot network area can also be a network area formed around a central network device, where the slave devices or ports connected to the central network device are considered as one area. Different central network devices form different network areas.

[0053] Step S20: Based on the preliminary traffic congestion indicators, apply a long short-term memory network to model the access prediction of hotspot areas to obtain a traffic prediction model; determine the probability of future traffic surges through the traffic prediction model.

[0054] The traffic prediction model described above integrates time period and user behavior data to simulate the potential impact of sudden events, thereby determining the probability of future traffic surges.

[0055] Step S30: If the probability of future traffic surge exceeds a preset threshold, obtain the processing capacity information of the remaining relay nodes from the network topology (processing capacity information is the relay node's ability to handle data traffic). Based on the relay node's processing capacity information and the probability of future traffic surge, group the remaining relay nodes using the K-means clustering algorithm to obtain multiple node grouping results (the node grouping results determine which relay nodes are suitable for allocation to specific hotspot areas to achieve load balancing; specific hotspot areas are nodes in subnets where multiple traffic clusters are located, corresponding to a traffic surge probability exceeding the threshold. In other words, the remaining relay nodes are assigned tasks to process the traffic information of nodes corresponding to traffic surges, because these nodes cannot handle the surged traffic information. Alternatively, the remaining relay nodes are allocated to nodes with surging traffic information to assist in processing the information of nodes corresponding to the surged traffic information. Figure 2 (as shown)

[0056] Step S40: Collect the available bandwidth and latency parameters of relay nodes in the grouping results of each node. Based on the available bandwidth and latency parameters of relay nodes and the probability of future traffic surges, perform traffic information allocation processing through particle swarm optimization combined with node state traffic impact analysis to obtain an optimized computing power allocation scheme.

[0057] It should be noted that the optimized computing power allocation scheme includes prioritizing the matching of high-processing-capacity nodes to high-congestion hotspot areas. In other words, high-processing-capacity relay nodes are prioritized to be matched with high-congestion (high traffic surge) nodes to process the traffic information corresponding to those nodes with even higher traffic surges.

[0058] Step S50: Based on the optimized computing power allocation scheme, dynamically adjust the subnet scheduling path, integrate the access prediction results to reroute traffic data flow to the allocated relay node, and determine the adjusted path stability score.

[0059] It should be noted that the above subnet scheduling path refers to the complete route taken by data traffic in the network topology from the source (e.g., the user request initiation point) through a series of intermediate nodes (including relay nodes) to finally reach the destination (hot nodes in hot areas). Through an optimized computing power allocation scheme, traffic will preferentially pass through relay nodes allocated to hot nodes, so that relay nodes can process traffic data, while reducing or eliminating relay nodes not allocated to hot nodes, thereby improving the efficiency of data processing.

[0060] The adjusted path stability score is calculated based on path availability, latency jitter penalty, packet loss rate penalty, and path consistency. Path availability measures the probability that a path can successfully transmit data within a specified time window. In the specific calculation, the system continuously sends probe packets (such as ICMP Ping) and calculates the ratio of total time to downtime. Downtime is determined by the response of probe packets. If multiple consecutive probe packets time out or fail to respond, it is considered a downtime. Finally, the availability value is (total time - downtime) divided by the total time, and the result is a value between 0 and 1. The higher the value, the more reliable the path.

[0061] The aforementioned latency jitter penalty is used to assess the degree of fluctuation in path latency. Greater fluctuation indicates a more unstable path. During calculation, the system first calculates the latency jitter value based on probe packets (usually expressed as the variance or standard deviation of latency). Then, this jitter value is compared to the system's preset maximum acceptable jitter threshold. The penalty is the ratio of the jitter value to the maximum threshold, but it cannot exceed 1 (i.e., if the jitter exceeds the threshold, the penalty is directly set to 1). Greater jitter results in a higher penalty value and a greater negative impact on stability. The packet loss rate penalty reflects the severity of data packet loss. The system calculates the packet loss rate based on the number of probe packets sent and acknowledged. (i.e., the number of unacknowledged packets divided by the total number of packets sent), and then compares this packet loss rate with the system's preset maximum acceptable packet loss rate threshold. The penalty value is the ratio of the packet loss rate to the maximum threshold, which also does not exceed 1. The higher the packet loss rate, the larger the penalty value, indicating that the path transmission quality is worse. Path consistency focuses on the stability of performance in multiple probes to avoid the problem of the path being "good and bad". In the calculation, the system calculates the coefficient of variation of the delay based on the delay data sequence of multiple probes (i.e., standard deviation divided by the mean), and then subtracts the coefficient of variation from 1. The smaller the coefficient of variation, the more stable the delay, and the closer the consistency value is to 1.

[0062] Then, by using the weighted sum of the path availability, latency jitter penalty, packet loss penalty, and path consistency, the adjusted path stability score is calculated. By comprehensively considering multiple dimensions such as availability, jitter, packet loss, and consistency, the adjusted path stability score fully captures the reliability and performance fluctuations of the path, avoiding the limitations of single-indicator evaluation. For example, even if the path availability is high, if the jitter is severe, the adjusted path stability score will be lower, thus more realistically reflecting the user experience. In addition, the adjusted path stability score, as a quantitative indicator, is compared with a preset threshold (e.g., 0.85) in step S60, directly triggering system operation. When the adjusted path stability score is lower than the threshold, the system automatically starts the optimization process (e.g., reallocating computing resources) without manual intervention, improving the efficiency of network management.

[0063] Step S60: If it is determined that the adjusted path stability score is lower than the preset stability threshold, the available bandwidth (available bandwidth or processing capacity for traffic data) of the remaining relay nodes is re-evaluated through an iterative optimization process, and the allocation scheme is updated to minimize the risk of traffic congestion, so as to obtain the final scheduling configuration.

[0064] Step S70: Based on the final scheduling configuration, deploy to the subnet to perform resource allocation and path rerouting operations in real time, and perform real-time monitoring and processing operations (real-time monitoring and processing operations monitor network performance indicators after execution to verify whether traffic congestion in hotspot areas has been alleviated).

[0065] It should be noted that the above-described embodiments of this application first collect two types of time-series data: historical traffic data and current access logs. Using time-series analysis methods, patterns are mined from the time dimension to identify hotspot areas (i.e., network areas with abnormal or persistently high traffic, such as application server clusters or database clusters experiencing abnormal traffic surges). Dynamic pattern characteristics reflecting these traffic changes (such as periodic peaks and trend increases) are then extracted. Finally, these characteristics are quantified into preliminary traffic congestion indicators, transforming the raw, chaotic, massive network data into structured indicators that can quantitatively assess network status. Further... In step S20, the initial traffic congestion index generated in S10 is received and modeled using a Long Short-Term Memory (LSTM) network. LSTM can not only learn short-term dependencies but also capture long-term patterns. By fusing time period (e.g., weekdays / weekends) and user behavior data, the model can simulate the potential impact of scenarios such as promotions and sudden events on traffic, outputting the probability of future traffic surges, thus achieving a leap from current situation perception to future prediction and proactively anticipating congestion risks. Furthermore, in step S30, a preset threshold is set as an action trigger. When the probability of traffic surges predicted in S20 exceeds this threshold, an action is taken. Intervention is necessary to obtain the processing capacity information of all remaining relay nodes from the network topology. Then, using the K-means clustering algorithm, nodes in the subnet are grouped according to their processing capacity, simplifying the large number of nodes into several resource pools with different processing capacity levels, preparing for the next step of precise matching. Furthermore, based on the node grouping results generated by S30, key performance indicators such as available bandwidth and latency parameters of nodes within each group are calculated. Based on the available bandwidth and latency parameters, an optimized computing power allocation scheme is generated. Through a priority matching principle, it is ensured that the most scarce resources (high-processing-capacity nodes) are used where they are most needed (high-performance nodes). (Congested areas), thereby theoretically maximizing resource utilization efficiency and achieving effective load balancing; further, based on the optimized computing power allocation scheme generated by S40, the subnet scheduling path is dynamically adjusted physically. Specifically, the access prediction results are integrated to reroute traffic data flow to the allocated relay nodes, actively determining the adjusted path stability score, assessing whether this path change introduces new risks (such as single point of failure, excessively long path, etc.), completing the actual redirection of traffic, transforming the prediction and allocation scheme into actual network state changes, and avoiding secondary problems of oscillation or instability caused by solving congestion through path stability;

[0066] Furthermore, in step S60, when the evaluation in S50 finds that the adjusted path stability score is lower than the preset stability threshold, it indicates that the initial allocation scheme is not ideal. Then, the iterative optimization process is initiated to re-evaluate the processing capacity of the nodes (which may be due to changes in network status) and update the allocation scheme, forming an internal closed-loop optimization mechanism to ensure that the final output scheme can not only alleviate congestion in theory, but also be stable and reliable in practice. Finally, the final scheduling configuration determined in S60 is deployed to the subnet for real-time execution, and real-time monitoring and processing operations are continuously performed, that is, monitoring the network performance indicators after execution and comparing them with the initial target to verify whether the traffic congestion in the hot spot area has been alleviated. The results of the entire analysis, prediction, and optimization process are truly applied to the production environment, and the monitoring results (new network performance data) are fed back to step S10 as part of the next round of historical traffic data.

[0067] Specifically, such as Figure 3 As shown, in step S40, based on the available bandwidth and latency parameters of the relay node and the probability of future traffic surges, traffic information allocation is performed through particle swarm optimization combined with node state traffic impact analysis to obtain an optimized computing power allocation scheme, including the following steps:

[0068] Step S41: Initialize particle swarm parameters, including determining hotspot nodes based on the probability of future traffic surges; establish a relay node set G based on all relay nodes, G={ , ,..., },in, For the i-th relay node (relay node) It also includes relay nodes. Available bandwidth Delay parameters of d and relay node i ); Establish a hotspot node set H based on all hotspot nodes, H = { , ,... },in, For the j-th hotspot node (hotspot node) It also includes hotspot nodes. (Probability of future traffic surge)

[0069] Delay parameters generally include propagation delay (the time required for data traffic to travel from the source to the destination in the physical medium), transmission delay (the time required to push all bits of a data packet onto the link, which usually depends on the data packet size and link bandwidth; the lower the bandwidth and the larger the packet, the higher the delay), processing delay (the time spent by network intermediate devices (such as routers and switches) checking data packet headers, determining the output link, checking for bit-level errors, etc.), and queuing delay (the time for data packets to wait in the router's output queue to be sent onto the link; typically, when network traffic approaches or exceeds the link's processing capacity, the queue begins to accumulate, and the queuing delay increases sharply). Simultaneously, the network topology is divided into multiple partitions based on the location of each node, with each partition containing multiple hotspot nodes and multiple relay nodes; the delay parameters of relay node i in this embodiment are... Specifically refers to processing delays;

[0070] Delay parameters c generally refers to round-trip time; round-trip time refers to the time from the source node (hotspot node) to the source node (hotspot node). Send a data packet to the target node (relay node) The round-trip time (RTT) reflects the total time required for the target node to immediately return a response packet to the source node. The RTT reflects the time it takes for a data packet to travel from the source node (hotspot node) to the target node. Sending a data packet through devices such as routers and switches to the target node (relay node) Round-trip time (RTT) is the time it takes for a data packet to travel from its source node to where it waits in the queue after passing through routers and switches, as well as other time-consuming processes. RTT reflects the most critical factor in network congestion. The more congested the network, the higher the queuing delay, and the greater the RTT.

[0071] It should be noted that the above-mentioned hotspot nodes refer to nodes whose probability of future traffic surge is higher than the preset threshold. In other words, the node will experience a surge in traffic at a preset time in the future. For example, if there are three nodes: work communication node, product order node, and communication node, and the product order node experiences a surge in traffic at a preset time (such as the date of an online shopping event like Singles' Day), then the product order node will be identified as a hotspot node.

[0072] The above-described embodiment of this application employs a hotspot-based integer encoding scheme: each particle represents a complete allocation scheme, encoded as a vector of length M, where the value j at the i-th position indicates the relay node. Assigned to hotspot nodes .

[0073] Step S42: Traverse each partition and perform processing: Establish an allocation matrix A based on the relay node set G and the hotspot node set H; each element aij in the allocation matrix represents the allocation of the i-th relay node... The computing resources are allocated to the j-th hotspot node. (The i-th relay node) Assigned to hotspot nodes Control relay node Perform auxiliary hotspot node (Perform traffic information processing); obtain multiple allocation result individuals based on each element aij in the allocation matrix A;

[0074] It should be noted that the above process of obtaining multiple allocation results based on each element aij in the allocation matrix A is a preliminary particle swarm initialization process. During this initialization, each relay node will only be assigned to one hotspot node. Each hotspot node will be allocated computing resources from one or more relay nodes. That is, a relay node can only be allocated to one hotspot node at a time. A hotspot node usually needs multiple relay nodes to share its surge in traffic.

[0075] Step S43: Calculate the initial fitness F based on the future traffic surge probability of each hotspot node in each assigned individual, the available bandwidth of each relay node, the latency parameters of the relay nodes, and the hotspot nodes.

[0076] The initial fitness F is calculated as follows:

[0077] ;

[0078] in, For reliability weighting coefficients, For delay weighting coefficients, These are the load balancing weighting coefficients, and the sum of the three is 1. Let be the probability of a future surge in traffic for the j-th hotspot node. d represents the available bandwidth of the i-th relay node. The number of relay nodes assigned to the j-th hotspot node. The network delay parameter value from relay node i to hotspot node j is obtained based on the network topology distance between the hotspot node and the relay node. The longer the topology distance, the higher the network delay parameter value. The calculation formula for the topology distance and the network delay parameter value will not be elaborated here. Let be the processing delay value of the i-th relay node. The total number of pairs to be allocated (i.e. The number of In other words, it refers to the number of allocation pairs formed by relay nodes being assigned to hotspot nodes. For example, if there is one hotspot node J, and three relay nodes I, II, and III, and all three relay nodes are assigned to hotspot node J, then the number of allocation pairs formed is... These are JI, J-II, and J-III respectively. The load rate of the i-th relay node in the set of relay nodes assigned to the j-th hotspot node. The average load rate of all relay nodes in the set of relay nodes assigned to the j-th hot spot node;

[0079] It should be noted that the above-mentioned initial fitness F is calculated by first traversing each hot spot node, multiplying the probability of a future traffic surge of the hot spot node by the total available bandwidth of all relay nodes it is allocated to, and then summing the results of this product of all hot spot nodes. The higher the probability of a traffic surge of a hot spot node, the greater its risk of congestion, and therefore it should receive more computing resources. Through the weighted summation of probability and resources, this component directly drives the optimization algorithm to tilt more bandwidth resources to the most dangerous and most needed places, thereby improving the overall reliability of the system in resisting congestion.

[0080] Next, the end-to-end latency of all "hotspot node-relay node" allocation pairs is calculated, including network transmission latency and relay node processing latency, and the average of these latency is taken. The latency component focuses on the transmission efficiency of data packets arriving at the auxiliary nodes. By minimizing the average latency, the algorithm tends to allocate high-quality relay nodes that are close to the hotspot nodes and have fast processing speeds, thereby ensuring that while sharing traffic, it does not significantly affect user experience or service real-time performance. Then, based on the estimated traffic, the expected load rate of each relay node is calculated, which is the ratio of estimated traffic to available bandwidth. Then, the variance of the load rate of all relay nodes is calculated. The smaller the load rate variance, the more balanced the load of all relay nodes. By penalizing high variance, the algorithm will promote a more even distribution of traffic across all available relay nodes, avoiding the unreasonable situation where a few nodes are overloaded while others are idle, thereby improving the robustness and resource utilization efficiency of the system.

[0081] Finally, these three components are combined according to preset weight coefficients to form a complete fitness function.

[0082] Step S44: Based on the initial optimization fitness F, multiple candidate computing power allocation schemes are obtained; the traffic surge caused by the probability of future traffic surges is analyzed to determine the traffic impact on hot nodes and relay nodes in the candidate computing power allocation schemes, and the load transition amplitude is obtained.

[0083] It should be noted that in the above embodiments of this application, the candidate computing power allocation schemes are selected based on the initial optimization fitness F, and the top n computing power allocation schemes in terms of fitness are retained as candidate computing power allocation schemes.

[0084] Step S45: The load transition amplitude is judged using a preset first load transition amplitude threshold and a second load transition amplitude threshold. When the load transition amplitude corresponding to the current candidate computing power allocation scheme is greater than or equal to the first load transition amplitude threshold, the candidate computing power allocation schemes corresponding to load transition amplitudes greater than or equal to the first load transition amplitude threshold and less than the second load transition amplitude threshold are updated according to a preset particle search step size to obtain a new candidate computing power allocation scheme. The new candidate computing power allocation scheme is then returned to the allocation result individual in step S43 above, and iterative processing is performed again until... The process continues until an optimized computing power allocation scheme is obtained. Candidate computing power allocation schemes whose load transition amplitude is greater than the second load transition amplitude threshold are eliminated because their load transition amplitude is too large, significantly exceeding expectations. Using them as the basis for particle optimization would inevitably lead to a more biased and incorrect candidate computing power allocation scheme for the entire population. When the load transition amplitude of the current candidate computing power allocation scheme is less than or equal to the first load transition amplitude threshold, the corresponding allocation scheme to be determined is output as the optimized computing power allocation scheme.

[0085] It should be noted that the above-described embodiments of this application first initialize the particle swarm parameter processing, accurately identify the "hot spots" that need assistance and the "relay node set" that can be scheduled based on the traffic prediction model, and clarify the key attributes of each node (available bandwidth, latency). Furthermore, by dividing the large-scale network topology into multiple partitions, the global optimization problem is decomposed into multiple more manageable local optimization problems, thereby reducing computational complexity.

[0086] Then, an initial candidate scheme population is generated. By generating an allocation matrix and multiple allocation result individuals, the first batch of candidate computing power allocation schemes are generated. The static efficiency of the scheme is evaluated by calculating the initial optimization fitness F, the merits of each candidate allocation scheme are quantitatively evaluated, and those obviously unreasonable schemes (such as misallocating resources to low-risk nodes) are quickly eliminated. The particle swarm is guided to conduct preliminary search and evolution in the direction of "high reliability, low latency, and load balancing".

[0087] Furthermore, for the high-quality candidate solutions selected in step S43, more rigorous dynamic stability stress tests are conducted. By simulating real-world scenarios of future traffic surges, the tests analyze whether the solutions will experience severe and unstable fluctuations under dynamic load impacts. The introduction of time dimension and impact analysis compensates for the shortcomings of the static evaluation in step S43 and identifies "fragile" solutions that are highly efficient in static terms but may cause service jitter due to drastic load changes in actual operation, thus providing a deeper insight for the final decision. Furthermore, based on the load transition amplitude calculated in step S44, a three-level processing strategy is implemented for the candidate solutions: the solution with the best stability is directly output as the final result; solutions with acceptable stability but room for improvement are retained and given a new search direction, re-entering the iteration process to find a better solution; and solutions with extremely poor stability are directly eliminated to prevent them from misleading the search direction of the entire population, ensuring that the final output computing power allocation solution has both high efficiency and stability.

[0088] Specifically, in step S44, the candidate computing power allocation scheme is analyzed in conjunction with the probability analysis of future traffic surges. The analysis of the traffic surge caused by the probability of future traffic surges will affect the traffic impact on hotspot nodes and relay nodes in the candidate computing power allocation scheme, and the load transition amplitude is obtained. This includes the following operation steps:

[0089] Step S441: Obtain the traffic surge time point based on the future traffic surge probability of each hotspot node; obtain the expected traffic corresponding to the future traffic surge probability of each hotspot node; determine the allocated carrying traffic of each relay node at each traffic surge time point based on the node grouping results of each relay node corresponding to each hotspot node in each candidate computing power allocation scheme (the allocated carrying traffic is the traffic that is allocated to the relay node for processing by the expected traffic of the hotspot node at the traffic surge time point); analyze the allocated carrying traffic with the available bandwidth and latency parameters of the relay node to obtain the load rate of the relay node at the traffic surge time point;

[0090] Step S442: Calculate the load change magnitude at traffic surge time point t and traffic surge time point t+1 for the i-th relay node; determine the load change rate of the relay node at traffic surge time point t based on the load change magnitude. ;

[0091] Step S443: Obtain the topological distance decay factor and traffic interaction intensity between the hotspot node and the i-th relay node in the current candidate computing power allocation scheme. ;

[0092] Step S444: Based on the load change rate and traffic interaction intensity The basic impact contribution is obtained by calculating the topological distance attenuation factor;

[0093] The calculation method for the basic impact contribution is as follows:

[0094] ;in, The topology distance attenuation factor is obtained from the topology distance analysis between the hotspot node and the i-th relay node;

[0095] It should be noted that the above-described embodiments of this application comprehensively calculate the load change rate of the relay node, the topology distance attenuation factor between the relay node and the hot spot node (i.e., obtained by analyzing after normalization of the topology distance; the longer the topology distance, the smaller the distance attenuation factor; the calculation method of calculating the topology distance attenuation factor by the topology distance between the hot spot node and the i-th relay node will not be elaborated here; or it can be obtained by matching the mapping relationship table of the topology distance and distance attenuation factor between the two nodes stored in a preset manner, which will not be elaborated here) and the traffic interaction intensity between the two to obtain the basic impact contribution.

[0096] Step S445: Calculate the load transition amplitude based on the basic impact contribution and the expected flow amplitude corresponding to the expected flow.

[0097] The calculation method for load transition amplitude is as follows:

[0098] ;

[0099] In the formula, For the t-th flow meter increment time point, The basic impact contribution of the i-th relay node to the j-th hotspot node at the t-th flow meter addition time point. The basic impact contribution of the i-th relay node to the j-th hotspot node at the (t+1)-th flow meter increment time point. Let be the expected traffic amplitude of the j-th hotspot node at time t. The time weighting factor from time point t to time point t+1;

[0100] It should be noted that the above-described embodiments of this application first construct a dynamic load model, obtain accurate traffic surge time points and expected traffic, and decompose these traffic into specific relay nodes according to candidate allocation schemes. Finally, the load rate of each relay node at each future key time point is calculated, and a refined load simulation environment that changes over time is constructed. This allows the server to go beyond static resource matching assessment and instead observe the dynamic carrying capacity of network resources under real, fluctuating traffic impacts, providing the most critical basic data for evaluating the stability and robustness of the scheme.

[0101] Furthermore, the node load volatility is quantified to assess the degree of drastic change in the load of each relay node and identify potential instability points in the network. Further, the inter-node correlation strength is evaluated by introducing the context of network topology and business logic to correct for the impact of load fluctuations in step S442, yielding topological distance (FJ, reflecting the physical or logical network path length) and traffic interaction intensity (FH, reflecting the frequency of historical or expected data interactions). By comprehensively calculating these three factors, a multi-dimensional weighted assessment of the impact is achieved. Finally, the overall load transition amplitude is calculated, providing a reference for the decision in step S45. For example, a higher load transition amplitude indicates that the candidate computing power allocation scheme will cause drastic and unstable fluctuations in the entire network load when dealing with future traffic surges, thus making it a high-risk scheme. Conversely, schemes with lower load transition amplitudes are considered more stable and reliable, realizing the transformation from prediction data to stability assessment, thereby enabling the selection of computing power allocation schemes that are not only efficient but also robust.

[0102] In the specific implementation process of the above-described embodiments of this application, the computing resources of relay nodes are allocated to hotspot nodes to assist hotspot nodes in processing traffic data, thereby reducing the traffic data pressure on hotspot nodes, improving their data processing capabilities, and averaging the server load to prevent computing power from being wasted. However, relay nodes may also experience their own data inflows. When these inflows are critical, such as those involving asset calculations, or when shopping nodes experience a surge in traffic during peak periods like Singles' Day, if relatively idle nodes like those for enterprise asset maintenance or server asset maintenance also receive important data packets, the computing power support provided by these relay nodes to hotspot nodes needs further analysis to avoid impacting their own traffic data processing.

[0103] Specifically, in step S60, the allocation scheme is updated to minimize the risk of traffic congestion, and the final scheduling configuration is obtained, including the allocation of computing resources for relay nodes and hot nodes, and the final scheduling configuration is obtained.

[0104] Specifically, the allocation of computing resources between relay nodes and hotspot nodes to obtain the final scheduling configuration includes the following steps:

[0105] Step S61: Traverse each relay node corresponding to each hotspot node, and set priority weights for each relay node according to the business processing type of the relay node.

[0106] It should be noted that in the above embodiments of this application, relay nodes that process critical assets are given higher priority weights, such as financial transactions and asset maintenance. By establishing their own business priority list, it is ensured that the business processing capabilities of high-priority relay nodes are guaranteed when allocating computing resources, and the normal operation of critical businesses is avoided due to the auxiliary hot nodes.

[0107] Step S62: Calculate the computing power allocation ratio of each relay node to the corresponding hot spot node based on the available bandwidth (available bandwidth is the remaining computing power of the relay node, that is, the data processing computing power remaining of the relay node after its own data processing), priority weight, and adjusted path stability score of each relay node.

[0108] It should be noted that in the above embodiments of this application, the calculation method for the computing power allocation ratio of each relay node to the corresponding hot spot node first considers three key factors: the available bandwidth of the relay node and the adjusted path stability score (i.e., the adjusted path stability score obtained in step S50). Specifically, for relay node i and hot spot node j, the weight calculation is based on the following logic:

[0109] Available bandwidth: The higher the available bandwidth of relay node i, the greater its potential to handle traffic. Therefore, it is given a higher weight in the allocation to make full use of its resources.

[0110] Adjusted path stability score (i.e., path stability value): The higher the adjusted path stability score from relay node i to hotspot node j, the more reliable the data transmission, and the higher the allocation weight, thus improving the overall service quality. After multiplying these factors, they are divided by the sum of the products of all nodes in the group to achieve normalization, ensuring that the sum of all weights in the group is 1, thus determining the resource allocation share of each relay node in the group (i.e., the computing power allocation ratio of each relay node to the corresponding hotspot node).

[0111] By combining available bandwidth, path stability, and priority weights, the system achieves refined allocation of computing resources, avoiding resource waste or overuse, improving overall utilization efficiency, and the introduction of adjusted path stability scores makes allocation more inclined towards stable paths, reducing the impact of network fluctuations on services and enhancing user experience.

[0112] Step S63: The server obtains a list of hot nodes as a whole, and calculates the priority scheduling evaluation value of each hot node based on the probability of future traffic surges, the business importance level of each hot node, and the real-time congestion level of the server.

[0113] It should be noted that when network resources are strained and there are multiple hotspot nodes, simultaneously allocating and processing traffic to each hotspot node will inevitably lead to traffic and computing power competition within the system, potentially causing server downtime. Therefore, when allocating surges in traffic to hotspot nodes, it is necessary to process them gradually to avoid server computing power explosion. Specifically, this involves calculating the priority scheduling evaluation value for each hotspot node. The calculation first considers three key factors: the probability of future traffic surges, the importance level of the services being carried, and the real-time congestion level. Specifically, for hotspot node j, its scheduling priority is based on the following logic:

[0114] The higher the probability of a future traffic surge, the greater the risk of congestion at that hotspot node in the future. Therefore, it is given a higher weight when calculating priority to reflect the principle of forward planning and guide the system to pre-emptively allocate resources to that node before congestion occurs, thus achieving proactive defense.

[0115] The business importance level of hotspot nodes is assessed based on the degree to which the business handled by the hotspot nodes affects the company's core value.

[0116] The server's real-time congestion level is based on real-time monitoring of overall server performance metrics such as CPU load, queue length, and response latency.

[0117] Finally, the three factors are weighted and summed to obtain the initial priority score of hotspot node j;

[0118] By comprehensively considering future risks, the value of business at hotspot nodes, and server congestion levels, we avoid the one-sidedness of single-dimensional decision-making and ensure that limited computing resources are always prioritized for nodes with the greatest impact on business, thereby maximizing the overall benefits of congestion mitigation.

[0119] Step S64: Obtain the expected traffic corresponding to the probability of future traffic surges, and perform the final allocation process based on the calculation results of steps 62-63 above (that is, combine the expected traffic with the computing power allocation ratio of each relay node to the corresponding hot spot node and the priority scheduling evaluation value of each hot spot node with the expected traffic corresponding to the probability of future traffic surges to determine the final scheduling configuration).

[0120] It should be noted that the above-described embodiments of this application first set the priority weight of relay nodes to identify the service processing type of relay nodes in the network, ensuring that the service flow of relay nodes is not sacrificed when supporting hot nodes; then, the computing power allocation ratio within the relay node group is calculated, and for each "hot node-relay node group" mapping relationship, the problem of "how to allocate within the group" is solved, comprehensively balancing three core elements: node capability (available bandwidth) and quality (path stability), to achieve fine-grained allocation of resources, explore and utilize the idle computing power of relay nodes, and improve resource utilization efficiency; furthermore, the priority scheduling evaluation value of hot nodes is calculated, when there are multiple hot nodes in the network... When there are multiple hot nodes and global resource constraints, the priority of resource allocation is determined. This is achieved by comprehensively assessing the future risk (probability of traffic surges), business value (importance level), and current congestion level of each hot node. A priority scheduling weight is calculated for all hot nodes to ensure that, during resource contention, servers can prioritize the most urgent and important hot nodes with limited computing resources. Finally, a final scheduling configuration is generated, combining the allocation ratio within the group, global scheduling priority, and specific traffic requirements (expected traffic). This determines how much computing power each relay node provides to the corresponding hot nodes and the order in which servers allocate computing power to support hot nodes. Figure 4 As shown, Figure 4 In the table, Ja, Jb, Jc, and Jd are the priority scheduling evaluation values ​​for each hotspot node, that is, from... Figure 4 The arrangement of hotspot nodes shown indicates the allocation of computing power to each hotspot node. In addition, Ia, Ib, Ic, Id, Ie, If, and Ig represent the computing power allocation ratio of each relay node to its corresponding hotspot node.

[0121] In summary, the network congestion mitigation method proposed in this invention for adjusting network resources first collects two types of time-series data: historical traffic data and current access logs. Using time-series analysis, patterns are extracted from the time dimension to identify hotspot areas (i.e., network areas with abnormal or persistently high traffic, such as application server clusters or database clusters experiencing abnormal traffic surges). Dynamic pattern features reflecting these traffic changes (such as periodic peaks or trending increases) are extracted from these hotspots. These features are then quantified into preliminary traffic congestion indicators, transforming raw, chaotic, massive amounts of network data into structured indicators that can quantitatively assess network status. Furthermore, by receiving the preliminary traffic congestion indicators and using a Long Short-Term Memory (LSTM) network for modeling, the model can learn not only short-term dependencies but also capture long-term patterns. By integrating time periods (e.g., weekdays / weekends) and user behavior data, the model can simulate the potential impact of scenarios such as promotions and sudden events on traffic, outputting the probability of future traffic surges. This achieves a leap from current situation perception to future prediction, proactively anticipating congestion risks. Finally, a preset threshold is set as an action trigger; when the predicted probability of a traffic surge exceeds this threshold... If the network topology is not clear, intervention is required. The processing capacity information of all remaining relay nodes is obtained, and K-means clustering is used to group the nodes in the subnet according to their processing capacity. This simplifies the large number of nodes into several resource pools with different processing capacity levels, preparing for precise matching in the next step. Furthermore, based on the node grouping results, key performance indicators such as available bandwidth and latency parameters of nodes within each group are calculated. Based on the available bandwidth and latency parameters, an optimized computing power allocation scheme is generated. Through a priority matching principle, it is ensured that the most scarce resources (high-processing-capacity nodes) are used where they are most needed. (High congestion areas), thereby theoretically maximizing resource utilization efficiency and achieving effective load balancing; furthermore, based on the optimized computing power allocation scheme, the subnet scheduling path is dynamically adjusted physically. Specifically, the access prediction results are integrated to reroute traffic data flow to the allocated relay nodes, actively determining the adjusted path stability score, assessing whether this path change introduces new risks (such as single point of failure, excessively long path, etc.), completing the actual redirection of traffic, transforming the prediction and allocation scheme into actual network state changes, and avoiding secondary problems of oscillation or instability brought about by solving congestion through path stability;

[0122] Furthermore, when the assessment finds that the adjusted path stability score is lower than the preset stability threshold, it indicates that the initial allocation scheme is not ideal. In this case, an iterative optimization process is initiated to reassess the processing capacity of the nodes (which may be due to changes in network status) and update the allocation scheme, forming an internal closed-loop optimization mechanism to ensure that the final output scheme can not only alleviate congestion in theory but also be stable and reliable in practice. Finally, the determined final scheduling configuration is deployed to the subnet for real-time execution, and real-time monitoring and processing operations are continuously performed. This involves monitoring the network performance indicators after execution and comparing them with the initial target to verify whether the traffic congestion in hotspot areas has been alleviated. The results of the entire analysis, prediction, and optimization process are truly applied to the production environment, and the monitoring results (new network performance data) are fed back to the initial step as part of the next round of historical traffic data.

[0123] In the specific execution process, the optimized computing power allocation scheme is also selected by calculating the load transition amplitude of each allocation scheme; specifically, the load transition amplitude is calculated by comprehensively considering the topological distance between hot nodes and the corresponding relay nodes, the intensity of traffic interaction, and the load change rate caused by traffic impact.

[0124] In addition, for the scheduling and configuration of computing resources when the server is congested, a comprehensive analysis is conducted by considering the business type, available bandwidth, and adjusted path stability score of the relay nodes, as well as the probability of traffic surges, business importance, and expected traffic of hot nodes. This analysis determines how much computing power each relay node should provide to the corresponding hot nodes and the order in which the server allocates computing power to support the hot nodes, thus avoiding server traffic congestion and ensuring that the allocation of computing resources is carried out in an orderly manner.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for network congestion mitigation of network resource adjustment, characterized in that, The following steps are included: By collecting historical traffic data and current access logs of each node in the subnet in real time, and using time series analysis methods, dynamic pattern characteristics of hotspot areas are extracted from historical traffic data and current access logs to obtain preliminary traffic congestion indicators. Based on preliminary traffic congestion indicators, a long short-term memory network is applied to model access prediction for hotspot areas, resulting in a traffic prediction model; the probability of future traffic surges is determined using the traffic prediction model. If the probability of a future traffic surge exceeds a preset threshold, the processing capacity information of the remaining relay nodes is obtained from the network topology. Based on the processing capacity information of the relay nodes and the probability of a future traffic surge, the remaining relay nodes are grouped using the K-means clustering algorithm to obtain multiple node grouping results. The available bandwidth and latency parameters of relay nodes in the grouping results of each node are collected. Based on the available bandwidth and latency parameters of relay nodes and the probability of future traffic surges, traffic information is allocated through particle swarm optimization combined with node state traffic impact analysis to obtain an optimized computing power allocation scheme. Based on the optimized computing power allocation scheme, the subnet scheduling path is dynamically adjusted, and the access prediction results are integrated to reroute the traffic data flow to the allocated relay node, and the stability score of the adjusted path is determined. If the adjusted path stability score is determined to be lower than the preset stability threshold, the available bandwidth of the remaining relay nodes will be reassessed through an iterative optimization process, and the allocation scheme will be updated to minimize the risk of traffic congestion, resulting in the final scheduling configuration. Based on the final scheduling configuration, it is deployed to the subnet to perform resource allocation and path rerouting operations in real time, and to perform real-time monitoring and processing operations.

2. The method of claim 1, wherein, Based on the available bandwidth and latency parameters of relay nodes, combined with the probability of future traffic surges, a traffic information allocation scheme is obtained through particle swarm optimization and node state traffic impact analysis. This scheme includes the following steps: Initialize particle swarm parameters, including determining hotspot nodes based on the probability of future traffic surges; establishing a relay node set based on all relay nodes; establishing a hotspot node set based on all hotspot nodes; and dividing the network topology into multiple partitions based on the location of each node, with each partition containing multiple hotspot nodes and multiple relay nodes. The execution processing is performed on each partition: a distribution matrix A is established according to the relay node set G and the hotspot node set H; each element aij in the distribution matrix represents that the computing power resource of the i-th relay node is allocated to the j-th hotspot node ; a plurality of distribution result individuals are obtained according to each element aij in the distribution matrix A; The initial fitness F is calculated based on the probability of future traffic surges for each hotspot node in each assigned individual, the available bandwidth and latency parameters of each relay node, and the network topology distance between the hotspot node and the relay node. Multiple candidate computing power allocation schemes are obtained by screening based on the initial optimization fitness F; the traffic surge caused by the probability of future traffic surges is analyzed for the candidate computing power allocation schemes, and the traffic impact on hot nodes and relay nodes in the candidate computing power allocation schemes is analyzed to obtain the load transition amplitude. An optimized computing power allocation scheme is obtained by iteratively filtering the load transition amplitude using a preset first load transition amplitude threshold and a second load transition amplitude threshold.

3. The method of claim 2, wherein, Delay parameter of the relay node i The processing delay value for the relay node.

4. The method of claim 3, wherein, Candidate computing power allocation schemes are selected based on the initial optimization fitness F, and the top n computing power allocation schemes in terms of fitness are retained as candidate computing power allocation schemes.

5. The method of claim 4, wherein, An optimized computing power allocation scheme is obtained by iteratively judging and filtering the load transition amplitude using a preset first load transition amplitude threshold and a second load transition amplitude threshold, including the following operation steps: When the load transition amplitude corresponding to the current candidate computing power allocation scheme is greater than or equal to the first load transition amplitude threshold, the candidate computing power allocation schemes corresponding to the load transition amplitude being greater than or equal to the first load transition amplitude threshold and less than the second load transition amplitude threshold are updated according to the preset particle search step size to obtain a new candidate computing power allocation scheme. The new candidate computing power allocation scheme is then returned to the above allocation result individual and iterated again until the optimized computing power allocation scheme is output. Candidate computing power allocation schemes whose load transition amplitude is greater than the second load transition amplitude threshold are filtered out. When the load transition amplitude corresponding to the current candidate computing power allocation scheme is less than the first load transition amplitude threshold, the corresponding allocation scheme to be determined is output as the optimized computing power allocation scheme.

6. The method of claim 5, wherein, The analysis of candidate computing power allocation schemes, combined with the probability analysis of future traffic surges, aims to analyze the traffic impact of these surges on hotspot nodes and relay nodes within the candidate schemes, thereby obtaining the load transition amplitude. This analysis includes the following steps: The traffic surge time points are obtained based on the future traffic surge probability of each hotspot node; the expected traffic corresponding to the future traffic surge probability of each hotspot node is obtained; and the allocated traffic capacity of each relay node at each traffic surge time point is determined based on the node grouping results of each relay node corresponding to each hotspot node in each candidate computing power allocation scheme. The load rate of relay nodes at traffic surge points is obtained by analyzing the allocated traffic and the available bandwidth and latency parameters of the relay nodes. Calculate the load change magnitude at traffic surge time point t and traffic surge time point t+1 for the i-th relay node; determine the load change rate of the relay node at traffic surge time point t based on the load change magnitude; Obtain the topological distance attenuation factor and traffic interaction intensity between hot nodes and the i-th relay node in the current candidate computing power allocation scheme; The basic impact contribution is calculated based on the load change rate, flow interaction intensity, and topological distance attenuation factor. The load transition amplitude is calculated based on the basic impact contribution and the expected flow amplitude corresponding to the expected flow.

7. The method of claim 1, wherein, The load jump amplitude is calculated by calculating the absolute value of the change in the base impact contribution of all relay nodes and hotspot nodes at consecutive time points, and then weighting and summing the expected traffic amplitude and the time weight factor The load jump amplitude is obtained.

8. The method of claim 7, wherein, The allocation scheme is updated to minimize the risk of traffic congestion, resulting in the final scheduling configuration, including the allocation of computing resources for relay nodes and hot nodes.

9. The method of claim 8, wherein, The allocation of computing resources between relay nodes and hotspot nodes is processed to obtain the final scheduling configuration, which includes the following steps: Iterate through each relay node corresponding to each hotspot node, and set priority weights for each relay node according to the business processing type of the relay node. The computing power allocation ratio of each relay node to the corresponding hot spot node is calculated based on the available bandwidth, priority weight, and adjusted path stability score of each relay node. The server obtains a list of hot nodes as a whole, and calculates the priority scheduling evaluation value of each hot node based on the probability of future traffic surges, the business importance level of each hot node, and the real-time congestion level of the server. The final scheduling configuration is determined based on the computing power allocation ratio of each relay node to the corresponding hot spot node, the priority scheduling evaluation value of each hot spot node, and the expected traffic corresponding to the probability of future traffic surges.

10. A storage medium, characterized by The storage medium stores a computer program, which, when executed by a processor, implements the steps of a network congestion mitigation method for adjusting network resources as described in any one of claims 1-9.

Images