云数据中心是云计算的基础支撑平台，提高资源管理效率和降低用户计算成本是其面临的重要挑战。项目首先依据云数据中心系统运行日志文件分析工作负载的日期与时间二维时间序列特征，提出基于二维长短期记忆神经网络的数据中心工作负载预测模型。其次，分析云数据中心物理资源能耗时序相关性，利用深度置信网络建立动态能耗模型和基于工作负载趋势预测的虚拟资源智能优化管理机制。第三，针对云应用程序任务优先约束依赖性和云数据中心存储系统的全局虚拟性，建立I/O数据感知的云应用程序动态DDAG任务模型，采用遗传算法与虚拟机融合技术实现系统能耗、性能多目标优化调度。最后，研究基于任务动态共享计费周期的云应用程序计算成本，提出任务子期限松弛化技术、成本优先涡流搜索调度策略及基于贪婪算法的任务调整方法。希能提高云数据中心资源管理效率和降低用户计算成本，为促进云数据中心应用效率提供一种可行方法。

Cloud data centers is the infrastructure platform for cloud computing, and its important challenges are to improve the efficiency of resource management and reduce user computing cost. Therefore, this project first investigates the workload date and time two-dimensional time series characteristics by cloud data centers system running log files. Then, we propose a data center workload prediction model based on two-dimensional long-term and short-term memory neural network. Third, we analyze the temporal correlation of physical resource energy consumption in cloud data centers, and use deep belief to establish the dynamical energy consumption model and the virtualization resources intelligence optimization mechanism based on workload trend prediction. Fourth, aiming at the dependence of cloud application task and the global characteristic of cloud data centers virtualization storage system, we build a cloud application I/O data aware dynamic DDAG task model, and we use genetic algorithm and virtual machine consolidation techniques to achieve a energy consumption, performance multi-objective optimization schleduling. Finally, we study cloud application computation cost based on task dynamic sharing billing period, and propose a task subdeadline relaxation technology, cost priority vortex search scheduling strategy, and task adjustment method based on greedy algorithm. It is hoped that this project can improve the efficiency of cloud data centers resource management and reduce user application computing cost, and provides a feasible way to promote the cloud data centers application efficiency.