Research On Task Optimization Mapping Algorithm Of Power System Scientific Calculations In Cloud Data Center

With the continuous expansion of multi-area interconnected power system and the gradual development of power cyber-physical system,the time complexity of various power scientific calculations related to power system optimization,operation and control,has greatly increased(e.g.,static stability analysis,power flow calculation,reactive power optimization and transient stability analysis).Meanwhile,the real time computing requirements are becoming higher and higher.However,the traditional power scientific computing acceleration methods are mostly carried out in a single server or a simple server cluster.Due to the limitation of computing resources,the acceleration performance is limited.Power cloud data center integrates computing,storage and communication ability.Based on virtualization and other technologies,it provides the possibility of real-time parallel and coordinated solution for scientific computing of large-scale multi-area interconnected grids.The mapping results from power scientific computing request to data center are key to the data center computing performance.Therefore,this paper studies the mapping from complex power scientific computing tasks to data center,and proposes a novel mapping method.It constructs a unified task directed acyclic graph and a minimum cut binary tree task mapping algorithm to accelerate the power scientific computing speed.Firstly,through the in-depth analysis of some important power science computing solution models,the common acceleration kernel,the decomposition coordination computing paradigm derived from the node admittance matrix,is extracted.And a new directed acyclic graph model corresponding to multi task computing communication equilibrium is constructed.Then,the characteristics of the task flow directed acyclic graph and the data center network architecture are analyzed.Considering the calculation time complexity of each node task in the task flow directed acyclic graph and the weight of data interaction between tasks,the proposed minimum cut binary tree algorithm is used to map the virtual machine where the key task is located to the physical machine first.Therefore,the completion time of the whole power scientific calculation is minimized.Finally,a variety of scientific computing hybrid operation experiments of IEEE 118 to super large-scale power system case 13659＿pegase are carried out on Cloudsim data center simulation platform.The results show that the proposed algorithm can improve the utilization rate of underlying physical machine resources and reduce the network throughput of data center,so as to obtain better computing acceleration performance.Therefore,it provides theoretical guidance and technical support for the real-time solution of the data center for complex power scientific calculation in large-scale power grid.