大数据驱动的异构无线网络用户体验质量QoE优化技术研究

Considering the distributed, dynamic and heterogeneous characteristics of Big Data in multi-source heterogeneous networks, we investigate the fundamental theories and key techniques of the processing of Big Data and QoE optimization in heterogeneous wireless networks. Firstly, we design and propose a dynamic network visualization of Big Data representation and pre-processing based on the adaptive selection of graph, matrix and tensor; Then, based on the analysis on Big Data from networking measurements, content, the information of space & time and the user's status and behavior, we propose and implement a three-stage evaluation and prediction process with training-clustering-classifying for QoE, with transforming the subjective into the objective, to obtain the automatic classification evaluation of QoE; Finally, we combine with routing, scheduling, interference managements to maximize QoE, through designing the cross-layer, large scale and distributed algorithm to find the optimal solution to QoE and the comprising Pareto or equilibrium solution between QoE and QoS. Finally, we achieve the optimization for user-centric network performance optimization for heterogeneous wireless networks. Meanwhile, we construct the processing testbed for Big Data based on the framework of Hadoop and optimization testbed based on VC++, to validate the proposed theories and algorithms.

针对多源异构网络大数据分布式、动态和异构等特点，研究网络大数据处理及用户体验质量优化的基础理论和关键技术。首先，设计并提出了基于图、矩阵和张量自适应选择的动态可视化网络大数据表征新方式和预处理新技术；其次，基于网络测量、业务内容、空时信息和用户状态及行为等数据分析，提出并实现基于实际测试-聚类-分类三阶段的QoE评价和预测过程，将用户状态和行为主观问题客观化，实现QoE的自动分类评测；最后，基于多源异构大数据的预测，以最大化QoE为目标，结合网络调度、路由选择和干扰管理等，设计了跨层的大规模分布式优化算法，求解大数据驱动的QoE优化解，并实现QoE和QoS的多目标帕累托或均衡折中分析，最终实现以用户为中心的异构无线网络性能评估与优化。同时，搭建基于Hadoop框架的多源异构网络大数据处理及其VC++优化系统的实验演示平台，验证所提理论及算法。

针对多源异构网络大数据分布式、动态和异构等特点，研究异构无线网络中大数据处理及大数据驱动网络用户体验质量QoE优化的基础理论和关键技术设计方法，项目从四个方面展开深入研究：.1）多源异构网络大数据分类聚类分析和挖掘（含数据表征和预处理），.2）多源异构网络空间大数据分析和定位，.3）多源异构网络大数据预测和增强学习优化，.4）多源异构网络大数据驱动的QoS和QoE优化分析。 .本项目的研究目标如下：建立适当的学习模型或随机模型，设计并实现优化的机器学习算法或随机分析方法，提取多源异构大数据关键特征，经过学习模型训练或随机理论推导，实现用户体验质量QoE的评价和预测；基于多源异构大数据的预测，设计并实现低复杂度分布式优化算法解决大规模分布式跨层优化问题，求解大数据驱动的Qos和QoE优化解，最终实现以用户为中心的异构无线网络性能优化。