【作者】 李杰；

【导师】 陈文宇；

【作者基本信息】 电子科技大学 ， 电子信息（专业学位）， 2024， 硕士

【摘要】 随着信息化时代的加速发展,通信网络的应用场景变得更加复杂和多样化,其中,特殊场景更是对通信保障和组网能力提出了前所未有的挑战。面临数据特殊且保密、通信层级结构互异、通信方式多样且混杂、通信状态易变且不稳定的场景迫切需要更高效、更可靠、更安全的通信网络解决方案。这不仅要求通信技术能够跨越自然和人为的障碍,实现无缝覆盖和高效传输,还需要网络设计能够灵活应对各种极端条件和复杂需求,确保在极端情况下能保持通信的畅通无阻。特殊场景通信网络属于超网络,相对于通用通信网络具有数据特殊、节点异质、多重链路以及指标复杂等特征。传统依赖于经典图论的算法模型不能准确完整的描绘此类网络的功能属性以及结构。这些特殊场景下所具有的复杂性特征是现有的网络分析方法无法有效处理的。针对上述问题,本文做了如下工作:(1)基于时序图神经网络,建立了特殊场景网络通信风险量化模型。结合数据的特殊性,进行专门的预处理与特征选取,通过图卷积网络和注意力机制的结合,构成时序图神经网络模型,为特殊场景通信网络中每一个节点提供风险等级预测,提前告知高风险节点的分布,使网络通信管理方提前做好应对方案。本文在特殊数据集上,使用不同模型测试对比了算法的预测准确率,相较于传统模型准确率提升了2.1%,实验结果证明了算法模型的有效性。(2)基于多目标优化技术,建立通信网组网优化模型。特殊场景网络通信中,两网络节点之间的连通路径可能是实时变更的,为满足实时性需求,不同时刻需要不同的路径连通方案。本模型基于通信节点的风险预估,通过综合考虑路径的抗毁性和便捷性,基于Pareto最优解集概念,给出多个较优组网方案供决策者选择,决策者可以根据不同的优先级和考虑因素,自主选择最适合的连接路径方案。在实验验证上,本文基于实际情况模拟了两大类的特殊场景,通过组网优化模型,均在毫秒级别内得到了较理想的网络节点路径连通方案,与传统单目标最短路径算法相比,性能和效果上有较大提升。更多还原

【Abstract】 With the accelerated development of the information age,the application scenarios of communication networks have become increasingly complex and diversified.In particular,special scenarios pose unprecedented challenges to communication assurance and networking capabilities.The urgent need for more efficient,reliable,and secure communication network solutions arises in scenarios where data is specialized and confidential,communication hierarchies are heterogeneous,communication methods are diverse and mixed,and communication states are volatile and unstable.This not only requires communication technology to overcome natural and artificial barriers to achieve seamless coverage and efficient transmission but also demands network designs that can flexibly adapt to various extreme conditions and complex demands,ensuring smooth communication even in extreme circumstances.Communication networks in special scenarios belong to hyper-networks and exhibit characteristics such as specialized data,heterogeneous nodes,multiple links,and complex metrics,as opposed to general communication networks.Traditional algorithm models based on classical graph theory cannot accurately depict the functional attributes and structures of such networks.The complexity features present in these special scenarios are beyond the effective handling capability of existing network analysis methods.In response to the above issues,this thesis undertakes the following work:1.Establish a special scenario network communication risk quantification model based on Temporal Graph Neural Networks.By combining the specificity of the data with specialized preprocessing and feature selection,a Temporal Graph Neural Network model is constructed through the integration of graph convolutional networks and attention mechanisms.This model provides risk level predictions for each node in special scenario communication networks,preemptively informing the distribution of high-risk nodes and enabling network communication managers to prepare response plans in advance.By testing different models on special datasets and comparing prediction accuracies,the algorithm in this thesis achieved a 2.1% improvement in accuracy over traditional models,demonstrating the effectiveness of the algorithm.2.Establish a communication network optimization model based on multi-objective optimization techniques.In special scenario network communication,the connectivity paths between two network nodes may change in real-time.To meet real-time requirements,different path connectivity schemes are needed at different times.Based on the risk estimation of communication nodes and considering the resilience and convenience of paths,this model utilizes the concept of Pareto optimal solution sets to provide multiple optimal networking solutions for decision-makers to choose from.Decision-makers can autonomously select the most suitable connection path scheme based on different priorities and considerations.Through the network optimization model,relatively ideal network node path connectivity schemes were obtained within milliseconds.Compared to traditional single-objective shortest path algorithms,significant performance and effectiveness improvements were achieved.更多还原