移动云中基于情景感知的服务应用可靠性保障机制研究

This project will use context-aware information combining with statistical learning theory and heuristic technology to study the service application reliability problem using lightweight technology in the mobile cloud environment, which will further develop the model optimization approach based on the incomplete information. First, for the selection problem of mobile service, we propose a service selection method based on genetic algorithm and minimal network latency technology, and then use the evaluation function with the environmental constrains to optimize the selection results. Then, we present the modeling and model optimization methods based on Hidden Markov Model (HMM) combing with time slices interception method for describing the relationship between the service state and running environment, and evaluate the correlation between the service state and running environment to solve the exception prediction problems. At the same time, for the fault diagnosis problem, we extend HMM to model the probability relation of service faults, service state streams and running environments, and then provide the best explanation for the current faults by decoding approach. Finally, the proposed approaches will be applied to the simulated mobile cloud system we will develop for verifying the effectiveness of our obtained results in this project.

本项目旨在应用情景感知信息，结合统计学习理论和启发式技术，研究移动云计算环境下的轻量级服务应用可靠性保障问题，并进一步发展基于不完备信息的模型优化方法。首先，针对移动服务选择问题，采用遗传算法和网络延迟最小化技术设计选择策略，进而利用带有环境约束的评价函数优化选择结果。然后，针对运行期间的服务环境异常采用隐马尔科夫模型结合时间片截取方法构建并优化服务状态与运行环境关系模型，通过评估当前服务状态与运行环境的匹配程度解决环境异常预测问题。同时，针对服务故障诊断方法扩展隐马尔科夫模型建模服务故障与服务状态流和运行环境间的概率关系，基于解码思想给出对于当前故障问题的最好解释。最后，将提出的方法应用到构建的移动云仿真系统当中，通过半实物仿真实验验证本项目提出方法的有效性。

在网络条件和终端资源可用性动态变化条件下确保移动服务的稳定运行成为目前移动云计算面临的主要挑战，针对该问题，本项目应用统计学理论和启发式技术，结合隐马尔可夫模型，研究轻量级的服务选择和服务异常预测问题，提出基于隐马尔可夫模型的服务选择和异常预测算法，以及相关模型的优化方法。本项目研究成果应用在移动云系统、推荐系统及高速贴片机系统中，取得了较好的应用价值，提高了服务应用运行的稳定性和可靠性。本项目资助出版专著1部，并获得2018年国家科学技术学术著作出版基金，发表期刊论文7篇，会议论文6篇，培养研究生10人，其中4人已毕业。