With the development of Internet technology, the centralized cloud computing cannot satisfy the real-time processing requirements of the edged terminal devices. Therefore, edge computing has emerged as a potential technology that can provide a bridge between data source and cloud computing centers. However, with the terminal devices of access network becoming more and more diversified, while providing convenience for people's life, the complicated network environment and heterogeneous network structure make the attack prediction and security defense become a new challenge in the edge computing environment. For example, (i) how to effectively measure the network parameters; (ii) how to reduce the occupied bandwidth caused by network measurement; and (iii) how to select valuable features based on measurement data. To address these problems, a topology simplification strategy based on the aggregation coefficient and the local information entropy of the network is proposed to reduce the bandwidth resource occupation caused by probe packets and the workload of network measurement. Moreover, in order to solve the problem that the data are out of date due to the long measurement period, an iterative strategy of environment-aware measurement interval is proposed. In addition, a feature selection strategy based on hybrid filtering method and chaotic genetic algorithm is proposed for flow classification. This method is more suitable for edge computing environments. Simulation experiments show that based on the open source KDD99 data set, compared with the other three methods, the strategy has better performance in multiple indicators such as precision, F-measure, false negative rate, and false positive rate. Unlike the previous work on classifiers, the three strategies we proposed complement each other and focus on the prework that has received less attention, which can make an important foundation for subsequent classification work.

中文翻译：

---

一种适用于软件定义边缘计算环境的稳健网络测量和特征选择策略

随着互联网技术的发展，集中式云计算已经不能满足边缘终端设备的实时处理需求。因此，边缘计算已经成为一种潜在的技术，可以在数据源和云计算中心之间架起一座桥梁。然而，随着接入网终端设备越来越多样化，在为人们生活提供便利的同时，复杂的网络环境和异构的网络结构使得攻击预测和安全防御成为边缘计算环境下的新挑战。例如，（i）如何有效地测量网络参数；(ii) 如何减少网络测量造成的带宽占用；(iii) 如何根据测量数据选择有价值的特征。为了解决这些问题，提出了一种基于聚合系数和网络局部信息熵的拓扑简化策略，以减少探测包对带宽资源的占用和网络测量的工作量。此外，为了解决测量周期长导致数据过时的问题，提出了环境感知测量间隔迭代策略。此外，针对流分类提出了一种基于混合滤波方法和混沌遗传算法的特征选择策略。这种方法更适合边缘计算环境。仿真实验表明，基于开源KDD99数据集，与其他三种方法相比，该策略在精度、F-measure、误报率、和误报率。与之前关于分类器的工作不同，我们提出的三种策略相辅相成，专注于不太受关注的prework，可以为后续的分类工作打下重要的基础。