Dynamic routing and bandwidth allocation method for wireless ad hoc network

Abstract

The invention discloses a dynamic route selection and bandwidth allocation method for a wireless ad hoc network, and relates to the technical field of photovoltaic power station dispatching control. The method aims to solve the problems of low power generation efficiency, fast assembly aging, poor power grid access stability and the like caused by the fact that an existing distribution method does not consider individual differences of photovoltaic strings, does not cooperate with energy storage optimization and is lagged in dynamic response. According to the method, multi-dimensional real-time data of a photovoltaic string, an energy storage system and power grid dispatching are collected, a multi-target optimization model is established, an improved particle swarm optimization algorithm is adopted for solving, cooperative power distribution of the photovoltaic string and the energy storage system is achieved, and real-time changes of power grid dispatching are adapted through a dynamic adjustment mechanism. The method can improve the total power generation efficiency of the photovoltaic power station, balance the operation state of components, stabilize the power fluctuation, meet the real-time scheduling requirements of a power grid, and prolong the service life of equipment.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic power plant scheduling and control technology, specifically, it relates to a dynamic routing selection and bandwidth allocation method for wireless ad hoc networks. Background Technology

[0002] Wireless ad hoc networks require no fixed infrastructure; nodes form a communication network through autonomous cooperation, offering advantages such as rapid deployment and strong resilience. However, these networks inherently possess characteristics such as high node mobility, dynamic topology changes, limited bandwidth resources, and heterogeneous service types, leading to numerous bottlenecks in traditional routing and bandwidth allocation schemes.

[0003] Static routing selection leads to poor adaptability: Traditional routing protocols (such as AODV and DSR) mostly make decisions based on path hop count or fixed link metrics, lacking dynamic awareness of link stability, node load, and service requirements. When node movement causes abrupt topology changes, route rediscovery latency is too long, easily leading to data loss or transmission interruption.

[0004] Bandwidth allocation is disconnected from routing selection: Existing solutions often adopt an independent decision-making model of "selecting routes first and then allocating bandwidth," where bandwidth allocation is based solely on the resource status of local nodes without considering the overall bandwidth bottleneck of the path. Even if a short path is selected, congestion may still occur due to insufficient bandwidth on critical links, reducing network throughput.

[0005] Insufficient QoS guarantee for services: Different services (such as real-time video, control commands, and data backup) have significantly different requirements for latency, bandwidth, and reliability. However, traditional solutions use a unified routing and bandwidth allocation strategy, which cannot meet the differentiated QoS requirements of heterogeneous services, resulting in high-priority services experiencing excessive latency or low-priority services preempting resources.

[0006] Imbalance between resource utilization and fairness: Nodes in distributed networks exhibit selfish competitive behavior. Traditional bandwidth allocation schemes either pursue the maximization of global throughput while ignoring node fairness, or lead to resource waste due to excessive fairness. At the same time, the lack of dynamic response to link fluctuations further exacerbates the problem of low resource utilization.

[0007] While existing improvement solutions attempt to introduce link quality awareness or bandwidth constraint mechanisms, they still have shortcomings: some solutions only focus on the shortest routing latency or bandwidth constraints, failing to achieve coordinated optimization of routing and bandwidth; some solutions adopt a centralized control architecture, which cannot adapt to the distributed characteristics of self-organizing networks and suffers from control latency; and some solutions fail to consider node privacy protection and the differentiated needs of heterogeneous services, resulting in limited practical application effectiveness. Therefore, there is an urgent need for an integrated solution that can achieve dynamic topology adaptation, routing and bandwidth coordination, and QoS differentiation guarantees.

[0008] In view of this, the present invention is proposed. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a dynamic routing and bandwidth allocation method for wireless ad hoc networks. By implementing cross-layer collaborative decision-making on routing and bandwidth, combined with intelligent sensing and optimization algorithms, it achieves efficient routing selection, fair bandwidth allocation, and differentiated QoS guarantees under dynamic topology, thereby improving network throughput, reducing transmission latency, and extending network lifetime.

[0010] The core idea of ​​this invention is: based on a closed-loop mechanism of "real-time perception - intelligent decision-making - collaborative optimization - dynamic adjustment," it acquires multi-dimensional network states through INT perception technology, implements distributed routing decisions using federated reinforcement learning, combines game theory to achieve fair and efficient bandwidth allocation, and finally achieves dynamic adaptation of routes and bandwidth through a cross-layer collaborative mechanism. The specific technical solution is as follows:

[0011] 1. Real-time network status sensing module

[0012] In-band Network Telemetry (INT) technology is used to embed metadata in service packets and collect network status indicators hop by hop with a collection period of 10-50ms to ensure real-time status awareness.

[0013] Perception metrics include link layer metrics (signal strength, link stability, packet loss rate, latency), node layer metrics (remaining energy, CPU load, cache utilization), and service layer metrics (service type, bandwidth requirements, latency threshold, priority).

[0014] Each node synchronizes its perceived local state to its neighboring nodes through a lightweight protocol, forming a local network state view, thus avoiding the delays and single-point-of-failure risks of centralized control.

[0015] 2. Federated Reinforcement Learning-Driven Dynamic Routing Module

[0016] Routing decision model construction: Using each node as an agent, a federated reinforcement learning model is constructed to achieve a balance between distributed routing decision-making and global optimization.

[0017] State space: contains key metrics such as link quality, node load, service requirements, and path hop count in the local network state view;

[0018] Action space: The set of next-hop neighbor nodes that the current node can choose;

[0019] Reward function: R = αR bw +βR lat +γR stab +δR energy , where αR bwβR is a reward for sufficient path bandwidth. lat As a reward for meeting latency targets, γR stab As a reward for link stability, δR energy The node energy balance reward is represented by α, β, γ, and δ, which are dynamically adjustable weight coefficients (which can be dynamically adjusted according to the network scenario).

[0020] Model training and updates: Each node trains the model based on local data, only uploading the model parameters (not the original data) to the aggregation node, updating the global model through the federated averaging algorithm, and then distributing the optimized model to each node, balancing model performance and data privacy protection.

[0021] Route generation and switching: The source node generates multiple candidate routes and calculates the comprehensive value of the paths based on the trained model; the route with the highest comprehensive value is selected as the primary route, while reserving 1-2 backup routes; when INT detects a decline in the quality of the primary route link (such as a packet loss rate exceeding 10% or latency exceeding the standard), it triggers a millisecond-level route switching to avoid transmission interruption.

[0022] 3. Game Theory-Optimized Dynamic Bandwidth Allocation Module

[0023] Service classification and demand mapping: Network services are divided into three levels: high priority (such as control commands, real-time video), medium priority (such as voice communication), and low priority (such as data backup). Different priority services correspond to different bandwidth demand thresholds and latency constraints.

[0024] Non-cooperative game model construction: The bandwidth allocation process is modeled as a non-cooperative game, with each node as a player in the game, aiming to maximize its own utility while taking into account global fairness.

[0025] Utility function: U i The bandwidth allocated to node i is , where is the service bandwidth requirement of node i, is the total link bandwidth, is the service priority weight, and B is the bandwidth allocated to node i. ireq B is the fairness coefficient. total w represents the total bandwidth of the link. i The business priority weight is λ, where λ is the commonality coefficient.

[0026] Nash equilibrium solution: Each node adjusts its bandwidth request strategy through a distributed search algorithm until a Nash equilibrium state is reached. At this point, any node can no longer improve its own utility by adjusting its strategy alone, thus achieving a balance between individual optimization and global fairness.

[0027] Dynamic bandwidth adjustment: Based on INT-aware link load changes, the bandwidth allocation results are updated every 20-50ms; when a high-priority service enters the network, bandwidth reallocation is triggered to prioritize its bandwidth requirements and then allocate the remaining bandwidth to medium and low-priority services.

[0028] 4. Routing-bandwidth cross-layer collaborative optimization mechanism

[0029] Collaborative decision-making trigger conditions: Collaborative optimization is triggered when any of the following conditions are met: network topology mutation (node ​​movement or failure), service type switching, link congestion (cache occupancy rate exceeds 70%), bandwidth allocation fairness deviation (maximum to minimum bandwidth ratio exceeds 3:1).

[0030] Collaborative optimization process: After collaboration is triggered, the routing module recalculates the bandwidth carrying capacity of candidate routes, and the bandwidth allocation module adjusts the allocation strategy based on the new routing topology; through cross-layer information exchange, it is ensured that route selection takes bandwidth constraints into account and bandwidth allocation adapts to routing characteristics, avoiding resource waste caused by independent decision-making.

[0031] Fast convergence mechanism: Introducing a link quality prediction model, which predicts link status changes within the next 100ms based on historical sensing data, and adjusts routing and bandwidth allocation schemes in advance to reduce the convergence time of dynamic adjustments.

[0032] Beneficial effects

[0033] Dynamically adapt to topology changes: Through INT millisecond-level awareness and federated reinforcement learning routing decisions, the routing switching latency is ≤5ms and the link failure recovery rate is ≥99%, effectively dealing with dynamic topology changes caused by node movement.

[0034] Differentiated QoS guarantee: Based on service classification and weight configuration, the bandwidth satisfaction rate of high-priority services is ≥95%, the latency is reduced by more than 30%, and there is no starvation phenomenon for low-priority services, realizing accurate QoS adaptation for heterogeneous services.

[0035] Balancing resource utilization and fairness: Through game theory-optimized bandwidth allocation, link bandwidth utilization is increased to over 85%, and the fairness deviation of node bandwidth allocation is ≤20%, taking into account both global throughput and individual fairness.

[0036] Low energy consumption and privacy protection: The routing selection takes into account the energy balance of nodes, reducing the average energy consumption of network nodes by more than 25%; the federated learning architecture avoids the transmission of raw data, ensuring the privacy and security of node data. Attached Figure Description

[0037] Figure 1 This is the logic diagram of the core algorithm of this invention. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will be combined with the embodiments of the present invention.

[0039] This embodiment takes a wireless ad hoc network in an emergency communication scenario as the application object. The network contains 50-100 mobile nodes with a node movement speed of ≤30km / h. The service types cover real-time video transmission (high priority), voice communication (medium priority), and emergency data backup (low priority). The total link bandwidth is 50-100Mbps.

[0040] Implementation steps

[0041] Network initialization: After each node starts up, it discovers neighbor nodes through INT technology, collects the initial network state (signal strength, remaining energy, initial load), and constructs a local network state view; the federated reinforcement learning model loads the initial parameters, and the game theory bandwidth allocation module initializes the weight coefficients and fairness coefficients.

[0042] Routing model training: Each node trains a reinforcement learning model locally based on locally collected state data and transmission feedback, and uploads the model parameters to the aggregation node every 5 minutes; the aggregation node updates the global model through the federated averaging algorithm and distributes it to each node to complete the model iteration.

[0043] Route selection execution: When the source node initiates a service transmission, it inputs service requirement parameters (type, bandwidth requirement, latency threshold), the model generates 3-5 candidate routes, calculates the comprehensive value of each route, selects the route with the highest comprehensive value as the main route, and synchronously informs all nodes on the path of the routing information.

[0044] Bandwidth allocation execution: The core nodes (link bandwidth bottleneck nodes) on the path initiate the game theory bandwidth allocation process. Each node, based on its own business needs and the total link bandwidth, solves the Nash equilibrium through a distributed algorithm to complete the initial bandwidth allocation; the allocation results are synchronized to all nodes on the path through control messages.

[0045] Collaborative dynamic adjustment: The INT module monitors the link status and service transmission quality in real time. When it detects that the link packet loss rate exceeds 10%, the latency exceeds the standard, or the service type is switched, it triggers collaborative optimization. The routing module switches to the backup route or regenerates the optimal route, and the bandwidth allocation module adjusts the bandwidth quota of each node in sync to ensure transmission continuity.

[0046] Continuous model optimization: Each node records feedback data such as route switching success rate, bandwidth utilization rate, and service QoS compliance rate, which are used to update model parameters and continuously improve decision-making accuracy.

[0047] Effect verification

[0048] In the application scenario of this embodiment, the network performance indicators are significantly improved compared to the traditional solution (AODV routing + static bandwidth allocation) after adopting the method of the present invention:

[0049] Network throughput: Increased by 40%-55%;

[0050] End-to-end latency: 35%-45% reduction for high-priority services, and 20%-30% reduction for medium- and low-priority services;

[0051] Packet delivery rate: increased to over 98%;

[0052] Network lifetime: Extended by 30%-40% (based on node energy balance optimization).

[0053] Core Creative Summary

[0054] A cross-layer collaborative architecture of "INT perception-federated reinforcement learning-game theory" is proposed, which realizes dynamic linkage between routing selection and bandwidth allocation for the first time, and solves the resource mismatch problem caused by independent decision-making in traditional solutions.

[0055] Federated reinforcement learning is introduced to achieve distributed routing decisions, balancing the model's global optimization capabilities with node privacy protection, and adapting to the distributed characteristics of wireless ad hoc networks.

[0056] The design incorporates a utility function that integrates business priority, bandwidth requirements, and fairness. By solving the problem using Nash equilibrium, it achieves a balance between differentiated QoS guarantees and fair resource allocation, adapting to heterogeneous business scenarios.

[0057] Based on INT technology, millisecond-level network status awareness and fast route switching are achieved, significantly improving network stability and transmission continuity in dynamic topology scenarios.

Claims

1. A dynamic routing and bandwidth allocation method for wireless ad hoc networks, characterized in that, Includes the following steps: S1: Real-time network status awareness: Using In-band Network Telemetry (INT) technology, metadata is embedded in service packets, and multi-dimensional status indicators of the link layer, node layer, and service layer are collected hop by hop. Each node synchronizes its local status to neighboring nodes to form a local network status view. S2: Federated reinforcement learning-driven dynamic routing selection: A federated reinforcement learning model is constructed with each node as an agent. Based on the local network state view, a candidate route set containing the main route and backup routes is generated through distributed training and global model aggregation. The optimal route is selected according to the comprehensive value of the path, and route switching is triggered when the link state is abnormal. S3: Dynamic bandwidth allocation optimized by game theory: Network services are prioritized and a non-cooperative game model is constructed. The Nash equilibrium is solved with the goal of maximizing node utility while taking into account global fairness. The initial bandwidth allocation is completed and dynamically adjusted based on changes in link load. S4: Cross-layer collaborative optimization of routing and bandwidth: When the preset triggering conditions are met, the routing selection module and the bandwidth allocation module are linked to adjust the routing topology and bandwidth quota synchronously, so as to realize the dynamic adaptation of routing characteristics and bandwidth constraints.

2. The method according to claim 1, characterized in that, In step S1, the multidimensional state indicators include: Link layer metrics: signal strength, link stability, packet loss rate, transmission delay; Node-level metrics: Node remaining energy, CPU load, cache utilization; Business layer metrics: service type, bandwidth requirement threshold, latency constraints, priority level; The acquisition period is 10-50ms, and the local state is synchronized through a lightweight protocol with a synchronization delay of ≤1ms.

3. The method according to claim 1, characterized in that, In step S2, the construction and training process of the federated reinforcement learning model includes: State space: contains key metrics such as link quality, node load, service requirements, and path hop count; Action space: The set of next-hop neighbor nodes that the current node can choose; Reward function: R = αR bw +βR lat +γR stab +δR energy , where αR bw βR is a reward for sufficient path bandwidth. lat As a reward for meeting latency targets, γR stab As a reward for link stability, δR energy The node energy balance reward is represented by α, β, γ, and δ, which are dynamically adjustable weighting coefficients. Model training: Each node completes model training based on local data, only uploading model parameters to the aggregation node. The global model is then updated using the federated averaging algorithm and distributed to each node, achieving a balance between privacy protection and global optimization.

4. The method according to claim 1, characterized in that, In step S2, the triggering conditions for route switching are: the packet loss rate of the main route link exceeds 10%, or the transmission delay exceeds the service delay constraint threshold, and the route switching delay is ≤5ms; the candidate route set contains 3-5 routes, and the link overlap rate between the backup route and the main route is ≤30%.

5. The method according to claim 1, characterized in that, In step S3, the service priorities are divided into three levels: high priority (control commands, real-time video), medium priority (voice communication), and low priority (data backup). Different priorities correspond to different bandwidth requirement thresholds and latency constraints. The utility function of the non-cooperative game model is: U i The bandwidth allocated to node i is , where is the service bandwidth requirement of node i, is the total link bandwidth, is the service priority weight, and B is the bandwidth allocated to node i. ireq B is the fairness coefficient. total w represents the total bandwidth of the link. i λ represents the business priority weight, and λ is the commonality coefficient.

6. The method according to claim 1, characterized in that, In step S3, the period for dynamic bandwidth adjustment is 20-50ms. When a high-priority service is accessed, bandwidth reallocation is triggered to prioritize the bandwidth requirements of high-priority services. The remaining bandwidth is then allocated to medium and low-priority services according to the utility function to avoid starvation of low-priority services.

7. The method according to claim 1, characterized in that, In step S4, the triggering conditions for the cross-layer collaborative optimization include: Network topology abrupt change: node movement leading to link disconnection or addition, or node failure; Changes in service status: service type switching, priority adjustment, or bandwidth requirement change; Link congestion: Node cache utilization exceeds 70%; Fairness bias: The maximum to minimum bandwidth allocation ratio between nodes exceeds 3:1; The collaborative optimization process also includes predicting link status changes within the next 100ms based on historical sensing data and adjusting routing and bandwidth schemes in advance.

8. The method according to any one of claims 1-7, characterized in that, The application scenarios of the wireless self-organizing network include emergency communication, intelligent transportation, and IoT edge networks. The number of network nodes is 50-100, the node movement speed is ≤30km / h, and the total link bandwidth is 50-100Mbps.

Images