近年来，随着通信与网络技术飞速发展，物联网在社会经济生活各个领域得到了广泛的应用，为人们的生产生活提供了很大便利。从工业物联网到智慧交通，从智能家居到智慧医疗，都是物联网的具体应用。丰富的应用场景、广阔的应用需求，推动着物联网设备数量急速增长，给物联网设备的管控和网络运营带来了挑战。物联网依托于互联网提供各类服务，由于物联网设备缺乏有效的隐私保护措施，导致隐私信息泄露事件频发，给用户的隐私权益带来很大威胁，研究物联网的流量隐私对解决隐私泄露问题有重要意义。然而现有研究缺乏大规模物联网流量数据隐私泄露检测方法。此外，物联网设备数量的增加也带来了大规模网络流量。做好网络流量规划管理，不仅影响着物联网设备的服务质量，也关系到互联网整体网络资源的统筹调配。对物联网设备流量建模分析，可以为流量管理提供有效的技术支撑。然而现有研究缺少结合物联网特点的流量预测工作。针对以上两个问题，本文通过对物联网流量开展模式分析和挖掘研究，论文创新点包括：1.提出了一种物联网流量隐私泄露检测方法：针对隐私泄露标注难的问题，基于物联网中最重要的三个实体用户、设备、平台，归纳隐私信息类别，提出自动化流量数据隐私标注方法。对物联网流量隐私泄露特点进行深入分析，提出了数据包量级的隐私特征生成和提取方法。使用5种经典机器学习分类方法，基于真实物联网流量数据实施隐私泄露检测，检测精确度达到97%以上。进一步，通过控制特征数量、数据规模、数据加密等变量，设计分析实验，检验方法的可靠程度，均取得很好性能。2.提出了一种基于物联网设备时序行为特征的流量预测方法。针对物联网设备流量促发性、间断性的特点，通过按时间间隔提取设备流量特征，并使用滑动窗技术处理，生成了可用的设备流量数据集。提出了基于物联网设备时序行为特征的流量预测方法。采用基于LSTM的自编码器，表征物联网平台、用户等因素对设备流量的影响，得到设备时序行为特征。将设备时序行为特征融入Transformer模型提升模型预测能力。实验证明，我们方法的性能较其它基线模型性能均有提升。
In recent years, with the rapid development of communication and network technology, Internet of Things(IoT) has been widely used in all aspects of social, economic and people’s life, such as Industrial IoT, Intelligent Transportation, Smart Home and Intelligent Health Care, etc. Multiple application scenarios and broad application requirements promote the rapid increase of the number of IoT devices, which brings challenges to the management of IoT devices and the operation of IoT networks. IoT provides services to users by connecting to the Internet. Because IoT devices lack effective privacy preservation tools, IoT privacy leakage frequently happens. People's privacy rights have been violated. However, there is still a lack of methods of large-scale IoT data privacy leakage detection. Meanwhile, the increase of IoT device numbers has also brought about massive network traffic. Good network traffic planning and management affect the service quality of Internet of Things devices and the overall coordination and deployment of Internet network resources. By modeling and analyzing the traffic of IoT devices, we can improve the service of IoT devices and provide practical technical support for network management. However, existing studies lack traffic prediction work that incorporates the characteristics of IoT. In response to the above two problems, this paper carries out pattern analysis and mining research on IoT traffic, and the innovation points of the paper include:1.We propose a detection method for IoT traffic privacy leakage. To solve the difficult privacy leakage labeling problem, we generalize the categories of IoT privacy leakage based on the three entities: user, device and platform in IoT. Then we design an automatic labeling method. We analyze IoT traffic privacy leakage characteristics and propose the method of privacy feature generation and selection. Using five classical machine learning methods, we evaluate our method on a real IoT traffic dataset. The detection accuracy achieve 97%. Further, analytical experiments are designed to test the reliability of the method by controlling variables such as the number of features, data size, and data encryption, all of which achieve good performance.2. We propose a traffic prediction method based on the temporal behavior features of IoT devices. For the characteristics of IoT device traffic that is proactive and intermittent, a usable device traffic dataset is generated by extracting device traffic features by time interval and processing them using the sliding window technique. An IoT device timing behavior feature extraction method is proposed, we use LSTM-based auto-encoder to characterize the impact of IoT platforms, users and other factors on device traffic. The device timing behavior characteristics are extracted. The device temporal behavior features are introduced into the Transformer model to enhance the modeling and prediction capability of the model on IoT device traffic. Experiments demonstrate that the performance of our method is improved over other baseline models.
