基于正交多项式神经网络的卫星网络QoS路由算法  

Satellite network QoS routing algorithm based on an orthogonal polynomials neural network

作　　者：刘贺语[1] 孙富春[1] 

机构地区：[1]清华大学计算机科学与技术系,智能技术与系统国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2013年第4期556-561,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金面上项目(60621062)

摘　　要：QoS路由问题作为多目标约束优化问题,是一种非确定多项式完全(NP-complete)问题,目前在地面网络中多使用启发式算法求解。卫星网络的拓扑时变特性要求路由算法快速收敛,而高昂的信息交换代价又要求尽量减少星间控制信息交换,这导致绝大多数探测导向型的启发式路由算法应用在星上时性能不高。该文引入一种基于正交多项式神经网络的卫星网络QoS路由算法,将数据包路由过程类比为在经训练过的神经网络中分类的过程;同时正交多项式的使用提高了训练速率,保证了拓扑周期内训练结果的有效性。仿真结果表明:该路由算法在满足用户QoS需求的同时还降低了链路拥塞、丢包率、呼叫阻塞率等指标。Quality-of Service （QoS） routing is a multi-objective optimization problem that is （nondeterministic polynomial complete） problem, which is normally solved by heuristic routing algorithms for terrestrial networks. The time-variabilities of satellite networks require that routing algorithms should converge rapidly while the high costs of information exchange requires minimization of the information exchange. The features of satellite networks thus impose harsh limits on most detection based heuristic routing algorithms on satellite networks. A satellite network QoS routing algorithm based on an orthogonal polynomials neural network was developed to assimilate the routing process into the classification process via a well trained neural network. The training velocities are enhanced by the orthogonal polynomials, which also guaranteed the effectiveness of the training results. Simulations show that this algorithm reduces the probabilities of link congestion, packet losses and call blocks while effectively satisfyingthe QoS requests from clients.

关 键 词：服务质量(QoS)路由 正交多项式 神经网络 卫星网络 

分 类 号：TP393.0[自动化与计算机技术—计算机应用技术]