网络用户行为分析及其预测技术研究

发布时间：2018-06-03 06:21

本文选题：神经网络 + 复杂网络　；参考：《北京邮电大学》2014年硕士论文

【摘要】：近几年,对下一代网络的研究已如火如荼,因为下一代网络所提出的架构理念能很好的解决现在互联网所存在的很多问题。与此同时,随着科技不断进步,用户对于所使用的网络及业务的服务质量要求不断提高。所以,立足于用户需求,再结合下一代网络的发展趋势,我们需要清楚而全面地把握网络用户的行为,预测网络用户行为的变化规律,从而为优化网络的性能和提高业务的服务质量提供一条路径。首先,本文介绍了下一代网络架构；总结了网络用户行为的研究现状；总结了神经网络用于用户行为分析及预测的研究现状。其次,为了进一步提高网络流量的预测精度,使模型能自适应不同的业务流量预测,我们研究了宽参数域下的回声状态神经网络算法(ESNs, Echo State Networks)。我们引入复杂网络理论以及基于生物侧抑制机制(LIM,Lateral Inhibition Mechanism)的思想提出了两种新型的回声状态网络算法： ·带有动态池预测的去耦合回声状态神经网络(DMESN+RP, Decoupled Mixed Echo State Network with Reservoir Prediction); ·带有最大信息量的DMESN(DMESN+Maxlnfo,Decoupled Mixed Echo State Network with Maximum Information)。与此同时,我们与传统的回声网络状态算法在预测精度,谱半径,参数鲁棒性等方面进行了仿真分析及对比。仿真发现我们所提出的DMESN+RP和DMESN+Maxlnfo在预测精度,谱半径参数变化范围及参数鲁棒性上要优于传统的回声状态神经网络。再次,我们将所提出的DMESN+RP和DMESN+Maxlnfo用于移动互联网的真实网络流量预测之中,从预测精度方面验证这种方案的实用性。最后,本文结合未来网络新型分层架构提出了一种基于DMESN+Maxlnfo的网络节点流量预测的新型网络链路分配策略。
[Abstract]:In recent years, the research on NGN has been in full swing, because the architecture of NGN can solve many problems existing in the Internet. At the same time, with the development of science and technology, the quality of service of the network and service is improved. Therefore, based on user needs and combined with the development trend of next generation network, we need to clearly and comprehensively grasp the behavior of network users and predict the changing law of network users' behavior. It provides a path for optimizing network performance and improving service quality. Firstly, this paper introduces the next generation network architecture, summarizes the research status of network user behavior, and summarizes the research status of neural network for user behavior analysis and prediction. Secondly, in order to further improve the accuracy of network traffic prediction and enable the model to adapt to different traffic prediction, we study the echo state neural network algorithm ESNs, Echo State networks in wide parameter domain. In this paper, we introduce the theory of complex network and the idea of LIMLlateral Inhibition Mechanism based on the biological side inhibition mechanism. We propose two new echo state network algorithms: De-coupled echo state neural network with dynamic cell prediction (DMESN RP, Decoupled Mixed Echo State Network with Reservoir prediction); DMESN(DMESN Maxlnfol decouped Mixed Echo State Network with Maximum Information with maximum amount of information. At the same time, the simulation analysis and comparison with the traditional echo network state algorithm in prediction accuracy, spectral radius and parameter robustness are carried out. Simulation results show that the proposed DMESN RP and DMESN Maxlnfo are superior to the conventional echo state neural networks in prediction accuracy, spectral radius parameter variation range and parameter robustness. Thirdly, we apply the proposed DMESN RP and DMESN Maxlnfo to the real network traffic prediction of mobile Internet, and verify the practicability of the proposed scheme in terms of prediction accuracy. Finally, this paper proposes a new network link allocation strategy based on DMESN Maxlnfo network node traffic prediction combined with the future network new hierarchical architecture.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】