Multi-route data transmission optimization method and system based on environmental feedback evolutionary game

Abstract

The application relates to the technical field of communication networks, and discloses a multi-route data transmission optimization method and system based on environment feedback evolutionary game, which comprises the following steps: constructing an undirected network graph, collecting router strategy state and global network environment state data; determining dynamic parameters based on the environment state data and constructing an interaction income matrix; determining income difference according to the income matrix, calculating strategy update probability through a Fermi function to randomly update the strategy, and then statistically obtaining the global average cooperation level; obtaining a candidate path set for the data flow to be transmitted, combining the average cooperation level of the routers along the path, the environment state and the congestion degree to obtain a comprehensive evaluation value through weighted evaluation; and selecting a target transmission path based on the evaluation value and transmitting data. The application establishes a bidirectional closed-loop feedback mechanism between micro routing strategies and macro environment resources, realizes self-adaptive load balancing, and effectively improves network transmission efficiency and overall resource utilization.

Description

Technical Field

[0001] This invention relates to the field of communication network technology, and in particular to a multi-route data transmission optimization method and system based on environmental feedback evolution game theory. Background Technology

[0002] With the continuous expansion of communication networks and the diversification of service types, data transmission faces increasingly complex network environments. Traditional routing protocols (such as OSPF and RIP) typically select paths based on static metrics (such as hop count and bandwidth) or local instantaneous states (such as queue length), lacking a coordinated consideration of dynamic changes in global network resources and long-term router behavior patterns, which can easily lead to local congestion, resource waste, or even network collapse.

[0003] While existing research has attempted to incorporate evolutionary game theory into routing strategy optimization, it generally suffers from the following key shortcomings: First, there is a lack of two-way closed-loop feedback between the game model and the network environment state. The router strategy only relies on local gains to evolve and cannot perceive changes in global resources, resulting in the evolution process being out of sync with the actual carrying capacity of the network. Secondly, the strategy evolution model and the routing decision mechanism are independent of each other, lacking a unified description of the dynamic system, making it difficult to analyze the evolution trend of network behavior from the perspective of system stability. Deeply integrating the "environment feedback evolution game model" with the "strategy-environment coupled dynamic system" requires solving a series of technical challenges, such as the asymmetric payoff structure under topological heterogeneity, the time-varying nonlinearity of environmental feedback, and the stability analysis of multivariable coupled systems. Existing technologies have not yet provided effective solutions.

[0004] Therefore, there is an urgent need for a multi-route data transmission optimization method that can integrate network topology heterogeneity, router behavior evolution, and environmental state feedback to achieve efficient utilization and long-term sustainability of network resources. Summary of the Invention

[0005] This invention aims to at least solve one of the technical problems existing in related technologies. To this end, this invention provides a multi-route data transmission optimization method and system based on environmental feedback evolutionary game theory.

[0006] The present invention provides the first technical solution.

[0007] A multi-route data transmission optimization method based on environmental feedback evolution game theory, comprising the following steps: Construct an undirected network graph of the communication network and collect policy status data of all routers in the network as well as global network environment status data; Dynamic parameters are determined based on the global network environment status data and two preset benefit coefficients, and a benefit matrix characterizing the benefits of interaction between routers is constructed using the dynamic parameters; the two preset benefit coefficients include a first benefit coefficient and a second benefit coefficient. The total revenue of each router is calculated based on the revenue matrix, the current policy of each router, and the policies of the neighboring routers. The revenue difference between choosing a cooperative policy and choosing a defective policy for each router is determined. Based on the revenue difference, the policy update probability of each router is determined by the Fermi function. The policy of each router is randomly updated according to the policy update probability. The global average level of cooperation in the network is calculated. The exponential term with the natural constant as the base in the Fermi function includes an exponential adjustment factor, which is input as the negative of the product of selection strength and payoff difference; wherein, selection strength is a preset parameter greater than zero, and payoff difference is the difference between the payoff of a node choosing a cooperation strategy and a betrayal strategy. For the data stream to be transmitted, obtain a set of candidate paths between the source router and the destination router, obtain the average cooperation level, data transmission congestion level and global environmental status of the routers along each candidate path, perform a weighted evaluation, and calculate the comprehensive evaluation value of each candidate path; Based on the comprehensive evaluation value of each candidate path, a target transmission path is selected from the candidate path set and data transmission is performed.

[0008] Furthermore, the steps for collecting the global network environment status data include: The system obtains the average cooperation level of routers in the network and updates the system's environmental state, which is mainly determined by the average cooperation level and coupling strength.

[0009] Furthermore, the steps for constructing the revenue matrix include: When both routers in the interaction choose the cooperation strategy, their respective gains are 1; when both choose the betrayal strategy, their respective gains are 0. When one party chooses a cooperation strategy and the other party chooses a betrayal strategy, the cooperating party receives 0 benefits, while the betraying party receives benefits equal to the dynamic parameters mentioned above.

[0010] Further, the step of randomly updating the policies of each router according to the policy update probability to calculate the global average cooperation level in the network includes: Each router, at each moment, determines whether to choose a cooperative strategy in the next moment based on its current payoff difference and the policy update probability, wherein the policy update probability is jointly determined by the payoff difference and the selection strength. Under weak selection conditions, the probability of each router choosing to cooperate is calculated using the Fermi function, and the policy state of each router is randomly updated based on the probability.

[0011] Furthermore, the step of calculating the global average level of cooperation in the network further includes: Calculate the average degree of the network across all routers in the network; Based on the average network degree, all routers are divided into low-degree routers and high-degree routers; The cooperation level of the low-degree routers and high-degree routers is obtained, which is jointly determined by the number of connections of the routers and the cooperation level of similar and dissimilar routers; the average cooperation level of the low-degree routers and high-degree routers is obtained based on the cooperation level.

[0012] Furthermore, in the step of obtaining the average cooperation level, data transmission congestion level, and global environmental state of the routers traversed on each candidate path and performing a weighted evaluation, the step of determining the average cooperation level of the routers traversed on each candidate path includes: Obtain the first number of low-degree routers and the second number of high-degree routers traversed along the candidate path; Based on the first and second quantities, a weighted average of the average cooperation levels of the low-degree routers and high-degree routers is calculated to obtain the average cooperation level of the routers passing through the candidate path.

[0013] Furthermore, in the step of obtaining the average cooperation level, data transmission congestion level, and global environmental state of the routers traversed on each candidate path and performing a weighted evaluation, the step of determining the data transmission congestion level includes: The queue lengths of all router ports traversed by the candidate path are collected, and the maximum or average queue length of the path is normalized and used as the data transmission congestion level.

[0014] Further, the step of selecting a target transmission path from the candidate path set and performing data transmission includes: The comprehensive evaluation values ​​of each candidate path in the candidate path set are sorted, and the candidate path with the largest comprehensive evaluation value is selected as the main transmission path, and the main transmission path is used as the target transmission path for data transmission. If it is determined that the maximum difference between the comprehensive evaluation values ​​of multiple candidate paths in the candidate path set is less than the difference threshold, then the multiple candidate paths are used together as the target transmission path, and a load balancing strategy is adopted to distribute the traffic of the data stream to be transmitted to the multiple candidate paths according to the proportion of the comprehensive evaluation value for data transmission. Following the step of transmitting data, the method further includes: The comprehensive evaluation value of the target transmission path is continuously monitored at a set period. When the comprehensive evaluation value of the target transmission path drops below the first threshold, or when there are other candidate paths in the candidate path set whose comprehensive evaluation values ​​are higher than the comprehensive evaluation value of the target transmission path and exceed the second threshold, a rerouting operation is triggered to switch the data transmission to the other candidate paths.

[0015] Furthermore, it includes a parameter adaptive adjustment step: Real-time monitoring of the global network environment state variables and the global average cooperation level; When the global network environment state variables and the global average cooperation level are detected to be continuously decreasing, the preset coupling strength is reduced, and the first benefit coefficient and the second benefit coefficient are adjusted to reduce the dynamic parameters in order to maintain the system within the desired stability domain.

[0016] Based on the first technical solution, a second technical solution is proposed.

[0017] A multi-route data transmission optimization system based on environmental feedback evolutionary game theory, used to implement the above-mentioned multi-route data transmission optimization method based on environmental feedback evolutionary game theory, includes: The topology modeling module is used to build and store network topology structures, and supports building the communication network into an undirected network graph. The data acquisition module is used to collect policy status and global network environment status data of all routers in the network; An environment status awareness module is used to collect network performance indicators in real time, and calculate and update the global network environment status variables based on the global network environment status data. The policy-environment coupled evolution module is used to calculate the benefit and benefit difference of each router based on the current global network environment state variables and router neighbor policies, determine the policy update probability of each router based on the Fermi function and complete the random update, and calculate the global average cooperation level and evaluate the system stability. The routing decision module is used to evaluate the quality of each candidate path based on the current global average cooperation level and the global network environment state variables, select the optimal target transmission path for data transmission, and support dynamic rerouting. The parameter adaptation module is used to monitor the changing trends of the global network environment state variables and the global average cooperation level to dynamically adjust the model parameters and ensure that the system operates within the desired stable range.

[0018] The above-described one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects: 1. It breaks through the limitations of traditional routing protocols that rely solely on static hop counts or local instantaneous queuing states, and creatively combines router behavior evolution with feedback from the macroscopic network environment, achieving closed-loop collaborative optimization of micro-policies and macro-environment.

[0019] 2. The degree classification modeling mechanism in the scheme can be well compatible with various complex topologies such as fully connected networks, scale-free networks, and small-world networks. It exhibits minimal performance degradation during large-scale network expansion and has excellent universality and robustness.

[0020] 3. It overcomes the rigid allocation problem of traditional protocols such as ECMP under burst traffic. By dynamically adjusting the routing path in real time based on the evolution status, it can intelligently avoid congestion, significantly reduce end-to-end latency and packet loss rate, and improve data transmission success rate.

[0021] 4. An innovative strategy – an environmental coupling fallback mechanism – can forcibly reduce the betrayal benefit when network resources are scarce, suppress malicious packet loss or delayed forwarding by routers, thereby delaying resource depletion and extending the healthy lifespan of the entire network.

[0022] 5. Based on evolutionary game theory and dynamical system theory, it provides a rigorous analysis of stability equilibrium points, giving the black-box network scheduling a clear theoretical mathematical explanation boundary, which facilitates parameter tuning by operation and maintenance personnel.

[0023] 6. The system is highly decoupled at each level (topology awareness, coupling evolution, routing decision, parameter adaptation), and can be directly deployed in existing SDN centralized controllers. It can even be used as an offline auxiliary engine and is compatible with traditional OSPF / BGP protocols, making engineering implementation less difficult.

[0024] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is a graph showing the evolution of cooperation level and environmental state in a heterogeneous network according to the present invention. Figure 3 This is a bar chart comparing the average end-to-end delay of the present invention with that of the traditional ECMP method; Figure 4 This is a comparison chart of the average packet loss rate of the present invention and the traditional ECMP method; Figure 5 This is a schematic diagram comparing the network throughput of the present invention with that of the traditional ECMP method; Figure 6 This chart compares the core performance metrics of the present invention with those of mainstream routing solutions (OSPF, Q-Learning, SDN+shortest path). Figure 6 Figure a shows a comparison of rerouting response times for different schemes. Figure 6 b is a comparison chart of bandwidth utilization for different schemes. Figure 6 c is a comparison chart of network throughput for different schemes. Figure 6 d is a comparison chart of the average packet loss rates of different schemes. Figure 6 e is a comparison chart of the average end-to-end delay of different schemes. Figure 6 f is a comparison chart of network jitter rates for different schemes; Figure 7 This is a comparison chart of the performance degradation of the present invention and the ECMP protocol under different network scales. Figure 7 Figure a shows a comparison of average end-to-end latency under different network sizes. Figure 7 b is a comparison chart of average packet loss rates under different network sizes. Figure 7 c is a comparison chart of bandwidth utilization under different network sizes. Figure 7 d is a comparison chart of network throughput under different network sizes. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. The following embodiments are used to illustrate this invention but cannot be used to limit the scope of this invention.

[0028] The technical solution of this invention aims to overcome the shortcomings of the prior art and provide a multi-route data transmission optimization method and system based on environmental feedback evolution game theory. By constructing a coupled evolution model of router behavior and network environment, dynamic adaptive routing selection and data transmission optimization are achieved, thereby improving the overall network performance and resource sustainability.

[0029] like Figure 1 The diagram illustrates the overall system architecture of the present invention.

[0030] like Figure 2 As shown, the evolution curves of cooperation level and environmental state in heterogeneous networks according to the present invention are illustrated.

[0031] This invention proposes a multi-route data transmission optimization method based on environmental feedback evolution game theory, comprising the following steps: 1. Construct a network topology model Collect topology information data of routers and terminal devices in the communication network, including router identifiers and their connections. Based on the collected data, construct an undirected network graph of the communication network in which the routers reside. ,in This represents a set of routers (routers or terminal devices). The set of edges (communication links) is represented by an adjacency matrix. Representation. Element Indicates router With router The connection relationships between them: Indicates router With router There is a connection. Indicates router With router No connection exists.

[0032] 2. Collect router status data At each discrete time Next, collect policy status data and global network environment status data from all routers in the network. Define the policy for a router to normally forward data packets as cooperation, represented as... Routers dropping or delaying the forwarding of data packets is indicated as... At the same time, define global network environment state variables. It is used to characterize the health of network resources. This indicates that network resources are sufficient and the network is in optimal condition. This indicates resource depletion and the worst-case scenario. This variable can be derived from multiple network performance metrics (such as average bandwidth utilization, end-to-end latency, packet loss rate, etc.) after normalization. Its state changes are influenced by the policy-environment coupling strength. The impact.

[0033] 3. Construct the payoff function and payoff matrix This step defines the revenue function between routers based on the current environmental state and the strategies adopted by each router, and constructs the corresponding revenue matrix to provide a basis for subsequent strategy updates.

[0034] First, we define the basic structure of the payoff matrix, specifically expressed as follows: When two routers interact, if both choose to cooperate, each receives a benefit of 1; if both choose to betray, each receives a benefit of 0. If one cooperates and the other betrays, the cooperating party receives 0, and the betraying party's benefit is a dynamic parameter. This parameter It is not a fixed value, but rather depends on the current environmental state. And two preset profit coefficients and The shared decision reflects the degree to which environmental changes incentivize betrayal. Among these, and satisfy or This causes the attractiveness of betrayal to reverse under different environmental conditions.

[0035] Dynamic parameters in the payoff matrix A direct mapping between environmental state and game payoffs is established, enabling routers to consider not only neighbor behavior but also implicitly perceive the global environmental resource state when updating policies. This design frees the evolutionary game model from being confined to the policy space, instead enabling bidirectional feedback between behavior and environmental resources through environmental variables, forming the microscopic foundation of a policy-environment coupled dynamic system. The nonlinear characteristics of this coupling mechanism and its stability analysis in multi-agent systems are the core technical challenges that distinguish this invention from traditional evolutionary game routing methods.

[0036] Based on this, the actual revenue of each router is calculated. A router's total revenue is the sum of its revenue from interactions with all neighboring routers. The specific calculation process needs to consider the router's own policy, the policies of its neighboring routers, and the connectivity between them. By summing the revenues of all neighboring routers and applying appropriate scaling, the overall revenue of that router under the current policy can be obtained.

[0037] Furthermore, to measure the motivation of routers to change their policies, the payoff difference between choosing a cooperative strategy and choosing a betrayal strategy is calculated. This payoff difference reflects the incremental benefit a router can gain by switching from betrayal to cooperation, and its magnitude depends on the number of connections the router itself has, the policy distribution of neighboring routers, and current dynamic parameters. The larger the payoff difference, the more inclined the router is to choose a cooperative strategy; conversely, the smaller the payoff difference, the more inclined it is to choose a betrayal strategy.

[0038] This difference in returns will serve as the core input for subsequent policy update mechanisms, driving the dynamic evolution of router behavior.

[0039] 4. Strategy Update Mechanism This step utilizes a stochastic policy update mechanism based on the Fermi function to simulate the dynamic process of routers adjusting their behavior according to payoff differences. At each time step, each router determines, with a certain probability, whether to choose a cooperative strategy for the next time step based on its current payoff difference. This probability is determined by both the payoff difference and the selection strength: the larger the payoff difference, the higher the probability of choosing cooperation; the selection strength controls the router's sensitivity to payoff differences, and the greater the strength, the more inclined the router is to choose the strategy with higher payoff.

[0040] Under weak selection conditions ( That is, when the impact of the payoff difference on strategy selection is small, the average probability of a router choosing cooperation can be approximated as a form that is linearly related to the average payoff difference.

[0041] Furthermore, a global average cooperation level is defined to measure the overall tendency of cooperative behavior across the entire network. This average cooperation level has a definite correspondence with the average probability of routers choosing to cooperate, thus linking micro-level individual policy updates with macro-level group cooperation levels.

[0042] Through the above mechanism, each router randomly adjusts its strategy based on its own benefit difference, thereby promoting the dynamic evolution of the overall network cooperation level.

[0043] 5. Simplified model under fully connected networks This step simplifies the policy-environment coupling dynamics model for fully connected network topologies.

[0044] In a fully connected network, each router is directly connected to all other routers, so each router's neighbor set is the set of all other routers in the entire network. Based on this symmetry, the local environments of all routers tend to be consistent, and the evolutionary behavior of the system can be fully described by two variables: the global average level of cooperation and the environmental state.

[0045] In such networks, the evolution of the average level of cooperation is jointly regulated by the current environmental state, router size, and revenue parameters. Specifically, the rate of change in the level of cooperation depends on the deviation between the current level of cooperation and the environmental state, as well as the amplification effect brought about by network size. The evolution of the environmental state is determined by the comparison between the current level of cooperation and the environmental carrying capacity: when the level of cooperation is below a certain threshold, the environment tends to improve; when the level of cooperation exceeds this threshold, the environment tends to deteriorate. The two are coupled together, forming a closed-loop feedback system.

[0046] By analyzing the equilibrium point and stability of this simplified model, we can reveal the basic laws governing the co-evolution of cooperative behavior and environmental state under fully connected networks, providing a theoretical benchmark for subsequent research on complex network topologies.

[0047] 6. Classification Modeling in Heterogeneous Networks For networks with highly heterogeneous router degree distributions (such as scale-free networks), this step proposes a modeling method based on router degree classification to characterize the impact of topological heterogeneity on policy evolution.

[0048] Based on the differences in router degree within the network, all routers are divided into two categories: routers with a degree less than or equal to the network average degree, and routers with a degree greater than the average degree. The former are typically more numerous but have fewer connections, while the latter are fewer but have denser connections, forming the hub routers in the network. For each type of router, its average cooperation level is defined, thus preserving topology information at a macroscopic level.

[0049] In heterogeneous network environments, the policy evolution of each router class is influenced not only by its own current level of cooperation but also by the level of cooperation of the other router class. This is due to the asymmetric connection patterns between routers of different degree classes: routers with higher degree classes are more likely to connect with routers with lower degree classes, and vice versa. Therefore, when constructing the evolutionary equation, it is necessary to introduce the connection probability between the two router classes to accurately reflect the impact of neighbor policy distribution on router revenue.

[0050] Specifically, the level of cooperation for each type of router is determined by its own number of connections, the cooperation levels of similar and dissimilar routers, and the overall environmental state. The evolution of the environment depends on the weighted average of the cooperation levels of the two types of routers (the weights being the proportion of each type of router in the total number of routers), the normalized sum of various router performance metrics, and the coupling strength. .

[0051] This classification modeling method can effectively reduce the complexity of the system while preserving the core characteristics of network heterogeneity, and provides a feasible theoretical framework for analyzing the emergence mechanism of cooperative behavior in heterogeneous networks and its co-evolution with the environmental state.

[0052] 7. Multi-route decision based on evolutionary state This step, based on the dynamic state information output by the aforementioned policy-environment coupled evolution model, designs a multi-path routing decision mechanism to achieve adaptive optimization of network transmission paths. The specific process is as follows: 7.1 Path Discovery and Candidate Set Construction: For any data stream to be transmitted, the source router first initiates a path discovery request. Relying on existing on-demand routing protocols (such as DSR, AODV, etc.), it probes and collects all reachable end-to-end paths between the source and destination routers, records the sequence of routers traversed by each path, and forms a candidate path set. ; 7.2 Path Quality Assessment: For each candidate path, a comprehensive evaluation value is constructed by considering the cooperation tendency of the routers it passes through and the overall environmental state. Specific indicators include: average cooperation level of the path. The global environment status and data transmission congestion level (maximum queue length or average queue length of the path) are considered. Based on the above indicators, the path evaluation value is defined as a weighted sum. The weight of each indicator can be dynamically adjusted according to the service type (such as low latency, high throughput, high reliability, etc.) to flexibly adapt to different service quality requirements.

[0053] 7.3 Path Selection and Traffic Distribution: Among all candidate paths, the path with the highest evaluation value is selected as the primary transmission path. If multiple paths have similar evaluation values, a load balancing strategy is used to distribute traffic across multiple paths to avoid single-point congestion and improve overall resource utilization. 7.4 Dynamic Adjustment and Rerouting: As network conditions (cooperation level and environmental conditions) evolve over time, this step continuously monitors path quality at fixed intervals or through event triggers. When the evaluation value of the current path drops below a preset threshold, or when a better path exists, the system proactively triggers rerouting to smoothly switch to the better path, ensuring that data transmission always operates in a superior state.

[0054] Through the aforementioned dynamic multi-route decision-making mechanism, the network can perceive and respond to changes in router behavior and resource status in real time, enabling route selection to have evolutionary and adaptive capabilities, thereby effectively improving transmission efficiency and network sustainability.

[0055] 8. Parameter adaptive adjustment The system can be configured with an adaptive mechanism to dynamically adjust key parameters (such as coupling strength) based on network performance. , revenue parameters and This is done to maintain the system within the desired stability region. For example, when a continuous decline in environmental conditions is detected, the system can be appropriately degraded. Alternatively, revenue parameters could be adjusted to encourage router cooperation.

[0056] System components: This invention also provides a multi-route data transmission optimization system based on environmental feedback evolution game theory, comprising the following modules: Topology modeling module: Used to build and store network topology structures (routers, links, degree distribution, etc.), supporting various network types such as fully connected, scale-free, and small-world.

[0057] Data acquisition module: used to collect policy status and global environment status data of all routers in the network; Environment Status Awareness Module: Collects network performance metrics (latency, packet loss rate, bandwidth utilization, etc.) in real time, calculates and updates the global environment status. (t).

[0058] Strategy-environment coupled evolution module: based on the current environment state The algorithm calculates the revenue and revenue difference for each router using the router neighbor policy, updates the router policy based on the Fermi function probability, and simultaneously calculates the global average cooperation level. The system equilibrium point and stability are analyzed, and the system evolution trend is predicted.

[0059] Routing decision module: based on the current and It evaluates the quality of candidate paths and selects the optimal path for data transmission; it also supports dynamic rerouting. This module can interact with existing routing protocols (such as OSPF and BGP) by modifying the routing table or adding policy routes.

[0060] Parameter adaptive module: dynamically adjusts model parameters based on system state (e.g., ... , and This ensures the system operates within the desired stable region. PID control, reinforcement learning, or rule bases can be employed.

[0061] Data storage and visualization module: Records historical data and provides functions such as evolutionary trajectory display and performance statistics.

[0062] The modules interact with each other through data interfaces and can be deployed on a centralized controller (such as an SDN controller) or run in a distributed manner on network routers.

[0063] Compared with the prior art, the present invention has the following beneficial effects: Global Cooperative Optimization: By combining the evolution of router behavior with network environment feedback, it achieves cooperative optimization of policies and environment, overcoming the limitation of traditional routing that only focuses on local instantaneous states.

[0064] Adapting to network heterogeneity: Supports fully connected networks and various heterogeneous network topologies (scale-free, small-world, random networks), and the model has universality and robustness.

[0065] Dynamic adaptive routing: It dynamically adjusts the routing path according to the real-time evolution status, effectively responding to network congestion and resource changes, and improving the transmission success rate and user experience.

[0066] Resource sustainability: Through policy-environment coupling mechanisms, excessive betrayal behavior is suppressed, network resource depletion is delayed, and network lifespan is extended.

[0067] Theoretical interpretability: Based on evolutionary game theory and dynamical system theory, it provides rigorous stability analysis, which facilitates network planning and parameter tuning.

[0068] Modular design: The system modules are decoupled, making them easy to extend and integrate into existing network architectures. For example, the routing decision module can be used alone in combination with traditional routing protocols, or offline simulation can be performed using the evolution analysis module.

[0069] The present invention will be further described in detail below with reference to specific embodiments.

[0070] Example 1: SDN-based Data Center Network Optimization 1. Implementation Scenarios and Hardware Environment Configuration This embodiment is applied to a medium-sized data center network, employing a fat-tree topology. This topology is commonly used in data centers, featuring high bandwidth, low latency, and strong scalability, thus meeting the optimization requirements of this invention for dynamic heterogeneous networks. Specific hardware and network scale configurations are as follows: (1) Network equipment: A total of 100 Layer 2 switches (as "routers" in this invention, responsible for data forwarding and policy decision response) are deployed. The model is H3C S5820V2-54QS-S. Each switch is equipped with 48 Gigabit Ethernet ports and 6 10 Gigabit SFP+ ports. The port buffer capacity is 16MB. It supports LLDP (Link Layer Discovery Protocol) and OpenFlow 1.3 protocol and can communicate seamlessly with the SDN controller. 20 servers (terminal devices) are deployed. The model is Dell PowerEdgeR750. Each server is equipped with 2 10 Gigabit network cards and is connected to different aggregation layer switches to achieve link redundancy. 1 SDN controller is deployed. The model is H3C CloudEngine S12700E. It is configured with a 16-core CPU, 64GB memory and 1TB solid-state drive. It serves as the centralized control core of the system and runs all modules of the system of this invention (except for some lightweight calculations of the policy update module).

[0071] (2) Topology details: The fat tree topology is divided into three layers: core layer, aggregation layer, and access layer. The core layer deploys 8 switches, the aggregation layer deploys 32 switches, and the access layer deploys 60 switches. Each switch in the core layer establishes a 10 Gigabit link connection with all aggregation layer switches, each aggregation layer switch establishes a gigabit link connection with 4 access layer switches, and each access layer switch establishes a gigabit link connection with 2-3 servers. The link bandwidth configuration of the entire network is as follows: 10 Gigabit (10Gbps) for the core layer-aggregation layer link, gigabit (1Gbps) for the aggregation layer-access layer link, and gigabit (1Gbps) for the access layer-server link. The default link transmission delay is 0.1ms (core layer-aggregation layer), 0.2ms (aggregation layer-access layer), and 0.1ms (access layer-server). The default link packet loss rate is 0.01% (under normal conditions).

[0072] (3) Network service scenario: Simulate typical data center services, including video streaming (high bandwidth, low latency requirements, bitrate of 10Mbps-100Mbps), data backup (high throughput, high reliability requirements, transmission rate of 50Mbps-200Mbps), and Web services (low latency, high concurrency requirements, data packet size of 1KB-10KB, concurrency of 1000-10000pps). The three services are deployed in a 3:4:3 ratio. The service is initiated through the server to continuously generate dynamic data streams, including burst traffic (the peak of burst traffic is 3-5 times that of normal traffic, the duration is 1-5 seconds, and it is randomly triggered) to test the optimization effect of the present invention under dynamic load.

[0073] 2. System Module Deployment and Initialization Configuration In this embodiment, the system adopts a centralized deployment mode. All modules run on the SDN controller, and the switches only handle data forwarding and status acquisition, without participating in complex policy calculations and evolution analysis. This reduces the hardware burden on the switches and improves system operating efficiency. The deployment details and initialization configuration of each module are as follows: (1) Topology Modeling Module: Deployed on the SDN controller, it automatically collects link information of the entire network via the LLDP protocol after startup, including parameters such as switch (router) identifiers, port identifiers, link connection relationships, link bandwidth, and link delay. The collection period is 100ms to ensure timely detection of link status changes (such as link failures and link recovery). After collection, an undirected network graph G(V,E) is constructed, where V is a set of 100 switches, identified as... E represents the link set, containing a total of 8×32 (core-aggregation) + 32×4 (aggregation-access) + 60×2.5 (access-server, average) = 256 + 128 + 150 = 534 links; the adjacency matrix A is a 100×100 matrix, with elements... When switch i and switch j have a direct link connection, It also records parameters such as link bandwidth and latency; when no direct connection exists, The topology modeling module stores the constructed topology graph and adjacency matrix in the SDN controller's database, updating it every 100ms. If a link failure is detected (e.g., ...), the module will update the database accordingly. If the value changes from 1 to 0, the topology is immediately updated and synchronized to all other modules.

[0074] (2) Data Acquisition Module: Deployed on the SDN controller, it establishes communication with each switch via the OpenFlow protocol to collect raw data of router (switch) policy status and global environment status at discrete time t. The acquisition period is 10ms (consistent with the acquisition period of the environment status awareness module). Among them, the router policy status is recorded through the virtual tags of the switches (without changing the switch hardware forwarding behavior, only used for policy evolution and routing decisions). The initial state is set to full cooperation, that is, the policy of all switches. The raw data for the global environment status includes port bandwidth utilization, port queue length, end-to-end transmission latency, and packet loss rate for each switch. Real-time data for each parameter is collected from all ports, and the average value is used as the corresponding indicator for that switch. The indicator data from all switches are then aggregated for subsequent calculation of the environment status 's'. The data acquisition module transmits the collected raw data in real-time to the environment status perception module and the policy-environment coupled evolution core module, while simultaneously storing it in the data storage and visualization module. The storage period is 1 second, and historical data is retained for 7 days for subsequent performance analysis and model optimization.

[0075] (3) Environment Status Awareness Module: Deployed on the SDN controller, it works in real time with the data acquisition module to receive the raw network performance indicators transmitted by the data acquisition module, and calculate and update the global environment status. The specific calculation process is as follows: The first step is to normalize each raw indicator to eliminate the influence of dimensions: ① Bandwidth utilization normalization: Let the average bandwidth utilization of all switch ports collected be u (unit: %), and the normalized value is ,in , The lower the bandwidth utilization, the larger the normalized value, indicating that the network resources are more abundant. ② End-to-end delay normalization: Let the average end-to-end delay collected be... (Unit: ms), delay threshold set to (Maximum acceptable latency for the data center network), the normalized value is ,when hour, ,when hour, The smaller the delay, the larger the normalized value; ③ Packet loss rate normalization: Let the collected average packet loss rate be l (unit: %), and the packet loss rate threshold be set to... The normalized value is ,when hour, ,when hour, The lower the packet loss rate, the larger the normalized value; ④ Queue length normalization: Let the average queue length of all collected switch ports be q (unit: packets), and set the queue length threshold to... The normalized value is ,when hour, ,when hour, The shorter the queue length, the larger the normalized value.

[0076] The second step is to calculate the global environment state s(t): using a weighted summation method, the calculation formula is as follows: in The average level of cooperation among routers. This is the normalized sum of the router's performance metrics (latency, loss rate, and throughput, etc.). For the strength of policy-environment coupling, The larger the value, the more environmentally damaging the router's behavior. The smaller the value, the weaker the impact of node behavior on resource status.

[0077] The weights of each normalized indicator are set according to business needs. For mixed services including video streaming, data backup, and web services, the weight allocation is as follows: bandwidth utilization (0.3), end-to-end latency (0.3), packet loss rate (0.2), and queue length (0.2). . This indicates that network resources are sufficient and in optimal condition (bandwidth utilization 0%, latency 0ms, packet loss rate 0%, queue length 0 packets). This indicates that network resources are exhausted and the network is in its worst condition (bandwidth utilization is 100%, latency is low). Packet loss rate Queue length The environment state perception module calculates s(t) every 10ms and transmits it in real time to the policy-environment coupled evolution core module, the routing decision module, and the parameter adaptation module.

[0078] (4) Strategy-Environment Coupled Evolution Core Module: Deployed on the SDN controller, it is the core module of the system of this invention, responsible for completing revenue calculation, strategy update, cooperation level output and stability analysis. The specific initialization configuration and operation process are as follows: ① Initialization parameter settings: Profit coefficient (less than 1) (Greater than 1), this parameter is set based on the following: when the environmental state s(t) is high (network resources are sufficient), ,at this time And follow As the value increases, the benefits of betrayal become higher, allowing for a small number of routers to betray (without affecting overall network performance); when the environmental state s(t) is low (network resources are scarce), ,at this time Furthermore, the coupling strength decreases as s(t) decreases, reducing the reward for betrayal and encouraging routers to choose cooperative strategies to avoid further resource depletion; (Used to control the degree to which router policies respond to environmental conditions), select the strength. (Weak selection condition, used to control the router's sensitivity to differences in revenue) The smaller the value, the smoother the router policy updates, avoiding network fluctuations caused by sudden policy changes; the policy update cycle is set to 100ms (consistent with the topology update cycle, avoiding computational burden caused by frequent updates).

[0079] The range of values ​​for the profitability coefficient is generally as follows: The specific value can be determined based on the number of routers. For example, in the embodiment given above: number of routers (<50), , .

[0080] ② Revenue Calculation Process: Every 10ms, receive s(t) transmitted from the environmental state awareness module and the router policy state transmitted from the data acquisition module, and calculate the total revenue and revenue difference for each router. First, based on... Calculate dynamic parameters Then, for each router Iterate through all its neighboring routers j (determined by the adjacency matrix A). (i.e., neighbors), computing router with neighbors Interactive benefits: If (Cooperation) and (Cooperation), then the interaction benefit is 1; if (Betrayal) and (Betrayal), then the interaction benefit is 0; if and ,or and Then the partner's profit is 0, and the betrayer's profit is... Next, the interaction gains of router i with all its neighbors are summed to obtain the router's... Total revenue Finally, calculate the router. The payoff difference between choosing a cooperation strategy and a betrayal strategy ,in It's a router. Always choose the total benefit when cooperating. It's a router. The total benefit of always choosing to betray. The larger the router Those who favor cooperation tend to choose a cooperative strategy, while those who favor betrayal tend to do so. For router The set of neighboring routers For router The strategy.

[0081] ③ Policy Update Process: Router policies are randomly updated every 100ms based on the Fermi function. Since the switch hardware does not change its forwarding behavior in this embodiment (to avoid affecting existing network operation), and the policy is only used as a virtual label for routing decisions, the policy update is only virtually calculated within the SDN controller. The expression for the Fermi function is: in, To select the strength.

[0082] ④ Calculation of Cooperation Level and Stability Analysis: After the strategy update is completed, the global average cooperation level is calculated. ( (N=100 is the number of routers) , The closer a value is to 1, the more cooperative behavior occurs in the network, and the better the network performance. Meanwhile, since the fat-tree topology in this embodiment is a heterogeneous network (the core layer, aggregation layer, and access layer switches have different degrees; the core layer switch degree is 32, the aggregation layer is 12, and the access layer is 5, with an average degree of...),... Therefore, based on the relationship between router degree and average degree, routers are divided into two categories: low-degree routers (degree ≤ 9.4, i.e., 60 access layer switches with a degree of 5) and high-degree routers (degree > 9.4, i.e., 8 core layer switches and 32 aggregation layer switches, totaling 40 switches, with degrees of 32 and 12 respectively). The average cooperation level of the two types of routers is then calculated. (Low) (Height), the construction includes , , The three-dimensional coupled dynamic system is analyzed to determine the system's equilibrium point and stability, predict the evolution trend of network cooperation level and environmental state, and identify when the system tends to become unstable (e.g., Continued decline If the value continues to decrease, a warning signal will be sent to the parameter adaptive module.

[0083] (5) Routing decision module: Deployed on the SDN controller, it is linked in real time with the policy-environment coupled evolution core module and the topology modeling module, based on the real-time evolution state ( , , The multi-path routing decision is implemented through the following process: ① Path Discovery and Candidate Set Construction: For any data stream to be transmitted (initiated by the server, containing information such as source IP, destination IP, service type, and transmission rate), the source router (access layer switch) initiates a path discovery request via DSR (Dynamic Source Routing Protocol). The DSR protocol uses a broadcast route request packet (RREQ) to probe all reachable end-to-end paths between the source router and the destination router. The route request packet contains information such as source IP, destination IP, and path records. Routers traversed by the path record will add their own identifiers. When the destination router receives the route request packet, it sends a route response packet (RREP) to return to the source router along the original path. The source router collects all reachable paths and records information such as the sequence of routers traversed by each path, link bandwidth, and link delay, forming a candidate path set. ,in In this embodiment, the number of reachable paths is... (Multipath characteristics of fat tree topology).

[0084] ② Path quality assessment: For each path in the candidate path set A comprehensive evaluation value is constructed by comprehensively considering three indicators: average cooperation level of the path, global environmental status, and path congestion level. This metric is used to measure the transmission quality of the path. The specific calculation and weighting of the metric are as follows: Average level of cooperation along the path : Calculation path The average cooperation level of all routers traversed; if the path passes through both low-degree and high-degree routers, then... ,in For path Number of low-speed routers passed through. This represents the number of routers traversed at different altitudes; if only a single type of router is traversed, then... or ; The larger the value, the stronger the willingness of routers along the path to cooperate, and the higher the reliability of forwarding. For low-level router cooperation, For a high level of router cooperation.

[0085] Path congestion The maximum queue length along a path is used as a congestion indicator when collecting path data. The maximum value is taken as the queue length of all router ports traversed. (Unit: packets), and then normalization is performed. ( ), The larger the value, the lower the path congestion level.

[0086] Comprehensive evaluation value calculation: The weights of each indicator are dynamically adjusted according to the business type. In this embodiment, the weight allocation for mixed businesses is as follows: average path cooperation level (0.4), global environment status (0.3), and path congestion level (0.3), i.e. If it is a single business, the weights can be adjusted. For example, for video streaming services (low latency requirements), the weight of path congestion can be increased to 0.4, the weight of average path cooperation level can be adjusted to 0.3, and the weight of global environment status can be kept at 0.3.

[0087] For different service types such as low latency, high throughput, and high reliability, the weight coefficients can be manually configured according to service quality requirements, or the optimal weight allocation can be adaptively calculated through algorithms based on service characteristics and real-time network status.

[0088] ③ Path selection and traffic allocation: A comprehensive evaluation of the candidate path set. Sort and select The longest path is used as the primary transmission path; if multiple paths exist... If the difference is less than 0.05 (approximately equal), a load balancing strategy is adopted to distribute the data stream according to the path. The traffic allocation ratio is distributed to these paths according to the following formula: the traffic allocation ratio is... (Only for) Paths with a difference of less than 0.05 are selected to avoid single-point congestion and improve overall resource utilization. For example, if two paths... , The difference is The traffic allocation ratio is , .

[0089] ④ Dynamic Adjustment and Rerouting: A dual monitoring mechanism of "periodic triggering + event triggering" is adopted to continuously monitor path quality. The periodic triggering period is set to 500ms, and the comprehensive evaluation value of each path is recalculated every 500ms. If the main transmission path If the value drops below a preset threshold (the threshold is set to 0.6, depending on network performance requirements), or if other paths exist... If the value is 0.1 or higher than the main transmission path, a rerouting is triggered; event triggers include link failures and path congestion caused by sudden traffic surges. ), Environmental conditions In cases where the signal drops below 0.3, or when the above events occur, rerouting is immediately triggered to smoothly switch to a better path (the switching process uses a gradual traffic change method to avoid packet loss, and the switching time is controlled within 10ms) to ensure that data transmission always runs in a better state.

[0090] (6) Parameter Adaptation Module: Deployed on the SDN controller, it receives data from the policy-environment coupling evolution module in real time. (Average level of cooperation) (Low router cooperation level) (High level of router cooperation), and the transmission of environmental state awareness modules. Dynamically adjust key model parameters (coupling strength) based on network operating status. Profitability coefficient , To maintain the system within the desired stability region (the stability region is defined as: , The specific adjustment rules are as follows: ① Coupling strength Adjustments: The initial value is 0.6, used to control the feedback weight of router policies on environmental states. When detected... Continuous decline (for 3 consecutive data collection cycles, i.e., 30ms), The decrease was ≥0.1), and Continuous decline (3 consecutive policy update cycles, i.e., 300ms, ε decrease magnitude) This indicates that network resources are strained and router tampering is increasing. In this case, it is appropriate to reduce the speed. To 0.4, reduce the negative feedback of environmental conditions on router policies; when detected Continuous increase (for 3 consecutive data collection cycles) (t) Increase ),and This indicates that the network condition is good, and the speed can be increased appropriately. Upgraded to version 0.8, enhancing the feedback of environmental conditions to router policies and improving resource utilization flexibility; when and At that time, maintain Unchanged. Through the above adjustments, As a regulating valve for system stability, it enables dynamic optimization of the policy-environment coupling strength.

[0091] ②Profitability coefficient , Adjustments: The initial value is 0.5. The initial value is 1.5, when When network resources are severely depleted, adjust , This further increases the impact of environmental conditions on the benefits of betrayal, making ,at this time The smaller the value, the smaller C(s), the lower the benefit of betrayal, and the more likely it is to force customized router cooperation; when When network resources are sufficient, adjust. , To reduce the impact of environmental conditions on the benefits of betrayal behavior, Allowing a small number of routers to betray their partners helps avoid resource waste caused by excessive cooperation; when At that time, maintain , constant.

[0092] (7) Data storage and visualization module: Deployed on the SDN controller, it uses a MySQL database to store historical data, including topology data, router policy status data, environmental status data, revenue data, cooperation level data, routing decision data, etc. The storage period is 1 second, and 7 days of historical data are retained; at the same time, a visualization tool (using Grafana) is integrated to display the network topology and global environmental status in real time. Change curves, global and average cooperation levels of the two types of routers , , The system tracks the evolution trajectory of the network, the comprehensive evaluation values ​​of each path and the traffic allocation, and the changing trends of network performance indicators (latency, packet loss rate, bandwidth utilization). It supports administrators in real-time monitoring of network operation status and allows the export of historical data for offline analysis to further optimize model parameters.

[0093] 3. Experimental verification and result analysis To verify the optimization effect of the method and system of this invention, a comparative experiment is set up in this embodiment to compare the method of this invention (denoted as EFG-MR, Environment Feedback Evolutionary Game-Multi Routing) with the traditional ECMP (Equal Cost Multi Path) routing protocol, and to test the network performance of the two methods under dynamic burst traffic. The specific experimental setup and results are as follows: Basic comparative experiments: EFG-MR vs. traditional ECMP Experimental parameter settings: The experiment lasted for 1 hour (3600 seconds), with the first 300 seconds being the normal traffic phase and the last 3300 seconds being the burst traffic phase (a burst traffic was triggered every 300 seconds, each burst lasting 3 seconds, and the peak burst traffic was 4 times that of the normal traffic); Test indicators included: average end-to-end network latency, average network packet loss rate, network throughput, and network resource utilization (bandwidth utilization); The experimental environment for both methods was completely identical, with only the routing decision mechanism differing. The ECMP protocol used hop-count-based load balancing and did not consider router behavior and environmental status feedback.

[0094] Experimental Results and Analysis: ① Average end-to-end network latency: The overall average latency of the ECMP protocol is 2.07ms; the overall average latency of the EFG-MR method of this invention is 1.49ms, a reduction of approximately 28% compared to ECMP. This is because this invention selects the optimal path based on router cooperation level and environmental conditions, effectively avoiding congested paths, and simultaneously reduces transmission latency by dynamically rerouting to switch to a better path in a timely manner. Figure 3 As shown.

[0095] ② Average packet loss rate: The overall average packet loss rate of the ECMP protocol is 0.41%; the overall average packet loss rate of the EFG-MR method of this invention is 0.24%, a reduction of approximately 40% compared to ECMP. This is because this invention, through a policy-environment coupling mechanism, encourages router cooperation, reducing packet loss, delayed forwarding, and other defective behaviors. Simultaneously, it distributes traffic through load balancing, avoiding packet loss caused by single-point congestion. Figure 4 As shown.

[0096] ③ Network throughput: The average throughput of the ECMP protocol is 314 Mbps, while the average throughput of the EFG-MR method of this invention is 376 Mbps, an improvement of approximately 20% compared to ECMP. This is because this invention optimizes path selection and traffic allocation, fully utilizes network resources, avoids resource waste, and ensures that data streams are always transmitted along the optimal path through dynamic rerouting, thus improving transmission efficiency. Figure 5 As shown.

[0097] Comparison Experiment of Mainstream Solutions: EFG-MR vs OSPF / Q-Learning / SDN + Shortest Path Experimental parameter settings: Continuing with the traffic rules and test duration of the basic experiment, new test indicators were added: rerouting response time and network jitter rate (measuring the degree of transmission delay fluctuation); the traditional OSPF protocol (link state), Q-Learning reinforcement learning routing, and SDN+shortest path routing were selected as the comparison group, and the EFG-MR of this invention was selected as the experimental group.

[0098] Experimental Results and Analysis: Table 1 below compares the core performance indicators of each solution. EFG-MR achieves the best performance across all core indicators, with the following specific advantages: ① Compared to the traditional OSPF protocol: latency is reduced by 21.6%, packet loss rate is reduced by 34.8%, throughput is increased by 15.9%, rerouting response time is reduced from 250ms to 10ms, an improvement of 96%, network jitter rate is reduced from 18.5% to 6.8%, solving the problem of OSPF's static measurement based solely on link bandwidth / hop count, and there is no obvious local congestion under burst traffic.

[0099] ② Compared to Q-Learning reinforcement learning routing: It does not require a large number of samples for training, reduces latency by 5.6%, reduces packet loss rate by 20%, increases throughput by 7.4%, and the rerouting response time is only 1 / 8 of that of Q-Learning. It also reduces network jitter rate by 44.7%, making it more suitable for dynamic and sudden network scenarios, and eliminating the training convergence delay problem of reinforcement learning.

[0100] ③ Compared to SDN+ shortest path routing: latency is reduced by 15.2%, packet loss rate is reduced by 31.8%, throughput is increased by 13.3%, and bandwidth utilization is increased by 10.8%. It avoids congestion problems caused by concentrated traffic on the shortest path and achieves balanced utilization of global resources.

[0101] Table 1. Comparison of core performance indicators between this invention and mainstream routing solutions.

[0102] Network Scale Robustness Testing: EFG-MR vs ECMP (50 / 100 / 200 routers) Experimental parameter settings: The number of routers was adjusted to 50 (small scale), 100 (medium scale), and 200 (large scale), while keeping the topology, service ratio, and burst traffic rules unchanged. The performance degradation of the two schemes under different scales was tested, with core test metrics consistent with the basic experiment. Test results are available in […]. Figure 7 It is displayed in the middle.

[0103] Experimental Results and Analysis: ① The performance degradation of EFG-MR is minimal under different network sizes. For small-scale networks (50 units), the average latency is 0.95ms, packet loss rate is 0.08%, throughput is 510Mbps, and bandwidth utilization is 85%. For medium-scale networks (100 units), the average latency is 3.02ms, packet loss rate is 0.60%, throughput is 1020Mbps, and bandwidth utilization is 82%. For large-scale networks (200 units), the average latency is 3.65ms, packet loss rate is 0.72%, throughput is 1980Mbps, and bandwidth utilization is 80%. Expanding the network size from 50 to 200 units only increases latency by 2.84 times, packet loss rate by 9 times, throughput increases almost linearly (3.88 times), bandwidth utilization remains above 80%, and network jitter is below 10%. The core reason is EFG-MR's heterogeneous network classification modeling and parameter adaptive mechanism, which can dynamically adjust the coupling strength θ and the benefit coefficients c0 / c1 according to the network size, adapting to changes in topological heterogeneity.

[0104] ② The ECMP protocol experiences a sharp performance degradation in large-scale networks. In a small-scale network (50 units), the average latency is 1.10ms, packet loss rate is 0.12%, throughput is 420Mbps, and bandwidth utilization is 78%. In a medium-scale network (100 units), the average latency is 4.20ms, packet loss rate is 1.00%, throughput is 850Mbps, and bandwidth utilization is 75%. In a large-scale network (200 units), the average latency is 7.85ms, packet loss rate is 2.15%, throughput is 1520Mbps, and bandwidth utilization is 68%. When the network scales up to 200 units, latency increases by 7.14 times, packet loss rate increases by 17.9 times, throughput only increases by 3.62 times, bandwidth utilization decreases by 12.8%, and network jitter exceeds 28%. Because ECMP lacks global environmental awareness, the topological heterogeneity and traffic dynamics in large-scale networks lead to the continuous spread of local congestion.

[0105] The embodiments verify the feasibility and effectiveness of the method and system of the present invention in data center networks. Through detailed deployment configuration, parameter settings and operation process, it is ensured that the present invention can be accurately implemented. At the same time, comparative experiments prove that the present invention can effectively reduce network latency, reduce packet loss rate, improve throughput and resource utilization, and enhance the adaptability and robustness of the network under dynamic load. It can be widely applied to various dynamic heterogeneous communication network scenarios.

[0106] like Figures 3 to 5The chart shown presents a performance comparison (latency, packet loss rate, throughput). This set of bar charts or time-series evolution graphs provides experimental control data. The data mapping shows that, compared to the traditional Equal Cost Multipath (ECMP) protocol, this technical solution significantly reduces the peak of the latency curve to 1.49ms under sudden load surges, a decrease of approximately 28%, while maintaining a low packet loss rate (0.24%), and significantly improving throughput by approximately 20%. This intuitively demonstrates that the environment-aware mechanism endows the network with proactive load balancing capabilities.

[0107] like Figure 6 and Figure 7 As shown, the comparison of core indicators and network scale robustness of multiple schemes is presented.

[0108] Figure 6 A comprehensive bar chart is used to compare this scheme with benchmarks such as OSPF and Q-Learning, demonstrating the advantages of this scheme in solving the problems of high convergence latency and blindness in static protocols in reinforcement learning.

[0109] Figure 7 The performance change is then expressed using network size (number of nodes expanding from 50 to 200) as the independent variable. This solution significantly improves network throughput while maintaining high bandwidth utilization after scaling, without a significant increase in average latency and packet loss rate. This strongly demonstrates the effectiveness of this invention in overcoming the curse of algorithmic dimensionality and achieving large-scale network expansion.

[0110] 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; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-route data transmission optimization method based on environmental feedback evolutionary game theory, characterized by the following steps: include: Construct an undirected network graph of the communication network and collect policy status data of all routers in the network as well as global network environment status data; Based on the global network environment status data and two preset benefit coefficients, dynamic parameters are determined, and a benefit matrix characterizing the interaction benefits between routers is constructed using the dynamic parameters. The two preset profit coefficients include a first profit coefficient and a second profit coefficient; The total revenue of each router is calculated based on the revenue matrix, the current policy of each router, and the policies of the neighboring routers. The revenue difference between choosing a cooperative policy and choosing a defective policy for each router is determined. Based on the revenue difference, the policy update probability of each router is determined by the Fermi function. The policy of each router is randomly updated according to the policy update probability. The global average level of cooperation in the network is calculated. The exponential term with the natural constant as the base in the Fermi function includes an exponential adjustment factor, which is input as the negative of the product of selection strength and payoff difference; wherein, selection strength is a preset parameter greater than zero, and payoff difference is the difference between the payoff of a node choosing a cooperation strategy and a betrayal strategy. For the data stream to be transmitted, obtain a set of candidate paths between the source router and the destination router, obtain the average cooperation level, data transmission congestion level and global environmental status of the routers along each candidate path, perform a weighted evaluation, and calculate the comprehensive evaluation value of each candidate path; Based on the comprehensive evaluation value of each candidate path, a target transmission path is selected from the candidate path set and data transmission is performed.

2. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, The steps for collecting the global network environment status data include: The system obtains the average cooperation level of routers in the network and updates the system's environmental state, which is determined by the average cooperation level and coupling strength.

3. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, The steps for constructing the return matrix include: When both routers in the interaction choose the cooperation strategy, their respective gains are 1; when both choose the betrayal strategy, their respective gains are 0. When one party chooses a cooperation strategy and the other party chooses a betrayal strategy, the cooperating party receives 0 benefits, while the betraying party receives benefits equal to the dynamic parameters mentioned above.

4. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, The step of randomly updating the policies of each router according to the policy update probability to calculate the global average cooperation level in the network includes: Each router, at each moment, determines whether to choose a cooperative strategy in the next moment based on its current payoff difference and the policy update probability, wherein the policy update probability is jointly determined by the payoff difference and the selection strength. Under weak selection conditions, the probability of each router choosing to cooperate is calculated using the Fermi function, and the policy state of each router is randomly updated based on the probability.

5. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, The step of calculating the global average level of cooperation in the network further includes: Calculate the average degree of the network across all routers in the network; Based on the average network degree, all routers are divided into low-degree routers and high-degree routers; The cooperation level of the low-degree routers and high-degree routers is obtained, which is jointly determined by the number of connections of the routers and the cooperation level of similar and dissimilar routers; the average cooperation level of the low-degree routers and high-degree routers is obtained based on the cooperation level.

6. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 5, characterized in that, The step of obtaining the average cooperation level, data transmission congestion level, and global environment status of routers along each candidate path and performing a weighted evaluation includes the following step: Obtain the first number of low-degree routers and the second number of high-degree routers traversed along the candidate path; Based on the first and second quantities, a weighted average of the average cooperation levels of the low-degree routers and the high-degree routers is calculated to obtain the average cooperation level of the routers passing through the candidate path.

7. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 5, characterized in that, In the step of obtaining the average cooperation level, data transmission congestion level, and global environment state of the routers along each candidate path and performing a weighted evaluation, the step of determining the data transmission congestion level includes: The queue lengths of all router ports traversed by the candidate path are collected, and the maximum or average queue length of the path is normalized and used as the data transmission congestion level.

8. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, The step of selecting a target transmission path from the candidate path set and performing data transmission includes: The comprehensive evaluation values ​​of each candidate path in the candidate path set are sorted, and the candidate path with the largest comprehensive evaluation value is selected as the main transmission path, and the main transmission path is used as the target transmission path for data transmission. If it is determined that the maximum difference between the comprehensive evaluation values ​​of multiple candidate paths in the candidate path set is less than the difference threshold, then the multiple candidate paths are used together as the target transmission path, and a load balancing strategy is adopted to distribute the traffic of the data stream to be transmitted to the multiple candidate paths according to the proportion of the comprehensive evaluation value for data transmission. Following the step of transmitting data, the method further includes: The comprehensive evaluation value of the target transmission path is continuously monitored at a set period. When the comprehensive evaluation value of the target transmission path drops below the first threshold, or when there are other candidate paths in the candidate path set whose comprehensive evaluation values ​​are higher than the comprehensive evaluation value of the target transmission path and exceed the second threshold, a rerouting operation is triggered to switch the data transmission to the other candidate paths.

9. The multi-route data transmission optimization method based on environmental feedback evolutionary game theory according to claim 1, characterized in that, Includes parameter adaptive adjustment steps: Real-time monitoring of the global network environment state variables and the global average cooperation level; When the global network environment state variables and the global average cooperation level are detected to be continuously decreasing, the preset coupling strength is reduced, and the first benefit coefficient and the second benefit coefficient are adjusted to reduce the dynamic parameters in order to maintain the system within the desired stability domain.

10. A multi-route data transmission optimization system based on environmental feedback evolutionary game theory, used to implement the multi-route data transmission optimization method based on environmental feedback evolutionary game theory as described in any one of claims 1-9, characterized in that, include: The topology modeling module is used to build and store network topology structures, and supports building the communication network into an undirected network graph. The data acquisition module is used to collect policy status and global network environment status data of all routers in the network; An environment status awareness module is used to collect network performance indicators in real time, and calculate and update the global network environment status variables based on the global network environment status data. The policy-environment coupled evolution module is used to calculate the benefit and benefit difference of each router based on the current global network environment state variables and router neighbor policies, determine the policy update probability of each router based on the Fermi function and complete the random update, and calculate the global average cooperation level and evaluate the system stability. The routing decision module is used to evaluate the quality of each candidate path based on the current global average cooperation level and the global network environment state variables, select the optimal target transmission path for data transmission, and support dynamic rerouting. The parameter adaptation module is used to monitor the changing trends of the global network environment state variables and the global average cooperation level to dynamically adjust the model parameters and ensure that the system operates within the desired stable range.

Images