随着无线技术的快速发展，许多机构都部署有大规模无线局域网。大规模无线局域网中不仅有大量网络设备，而且还有复杂的部署环境和大量的用户，这为无线网络的管理带来了很多挑战。（1）无线网络运行管理难度高，许多大型无线网管理仍旧依赖于网络管理人员的经验，但复杂的网络环境已经很难仅凭经验来全面理解，而是需要测量和数据的支持。（2）无线网络性能优化难度高。一方面，性能受位置和设备类型的影响，另一方面，不同的应用关注的性能指标也不一样。这使得管理者很难用简单的标准衡量网络运行情况。（3）用户行为也会对无线网性能产生影响，对用户行为的研究有助于网络的管理和配置的个性化。但由于可取得的数据有限且用户行为往往具有高度复杂性，用户行为不能用简单的模型进行建模。为了解决以上这些挑战，本文以清华大学无线校园网为对象进行了研究，主要的研究和贡献如下。（1）本文以清华大学无线网为对象设计了无线网测量分析平台。通过被动测量的方法从无线网中获取了大量的运行日志并将这些日志加以整合。对于难以准确获取的信息使用多个数据集联合验证进行识别，对于没有直接记录的数据采用近似的方法获取，能够支持应用性能和用户移动性分析。（2）本文从应用角度对不同类型应用的流特征和各自关注的性能指标开展了大规模测量。研究发现不仅不同类型应用有不同的流特征，用户位置和设备类型也会对流特征产生明显影响。通过使用相对信息增益衡量不同因素对无线网络性能的影响，发现不同场合下明显影响无线网络性能的因素，并对这些因素影响性能的原因进行了分析。（3）本文对用户移动性进行了研究。现有的模型没有解决用户移动行为数据收集过程中难以避免的错漏问题，也没有对用户选择路径时的偏好和习惯进行捕捉。针对这些问题，我们提出了基于图卷积和注意力机制的用户移动行为预测模型。我们在三个数据集上对该模型进行了对比实验，包括两个公开数据集和一个收集自清华大学无线校园网的数据集?

With the rapid development of wireless technology, many organizations have deployed large-scale Wireless Local Area Networks (WLANs). The great quantity of network devices and users, along with complex network environment, brings many challenges to network management. (1) Managing wireless networks is challenging. Many network administrators manage network only by experience, which cannot deal with the complex environment in large-scale WLANs. Therefore, measurement and statistics of WLANs are in need to understand large-scale WLANs. (2) Optimizing wireless network performance is challenging. On the one hand, location and device type can affect performance. On the other hand, users are sensitive to different performance metrics when they use different applications. Hence the network performance of large-scale WLANs cannot be measured by a single metric. (3) User behaviors also affect network performance. Research on user behaviors can facilitate the measurement and configurations of networks. But modeling user behaviors is challenging due to the lack of data and the complexity of user behaviors. To address these problems, we conduct researches based on campus network of Tsinghua University, and the main content and contributions are summarized as follows.(1) We design and implement a wireless network measurement and analysis system based on campus network of Tsinghua University. We collect and integrate multi-source datasets from campus network by passive measurement. For the information suffering from inaccuracy, we use multi-source datasets to verify it. And for the information not recorded directly, we use related data to make approximation. The collected information can support following application performance analysis and user mobility research.(2) We conduct a large-scale measurement on application-level traffic patterns and performance. We find that flow patterns vary with not only application types, but also user locations and device types. By evaluating the influence of factors on performance using relative information gain, we find crucial factors that affect performance. Meanwhile, we analyse the reasons for the influence of these factors.(3) We make research on user mobility prediction and modeling. Existing models have problems to solve the sparsity and inaccuracy of trajectory data and they also neglect the preference and habits when users choose future paths. To solve these problems, we propose a novel model based on graph convolution and attention mechanism. To evaluate the effectiveness of our model, we conduct experiments on three real-world datasets, including two open datasets and one collected from campus network of Tsinghua University. The results show that our model outperforms state-of-the-art baselines.