Research On Scheduling Strategy And Parallel Load Flow Solution In Spark

Cloud computing is a key issue of power system research.The parallel computing framework for data transmission using Spark clusters in power flow analysis has becoming a hot topic.Compared to traditional serial systems,Spark Cloud can reduce the convergence time of operations.For data-intensive tasks,it can increase the running speed by dozens of times.While the parallel computing of power system is developing rapidly,how to improve the performance and memory utilization of Spark scheduling system has becoming an urgent problem to be solved.This thesis is based on the cloud computing engine Spark.Firstly,it studies the behavior of RDD,optimizes the top-level scheduling mechanism such as task filtering and partition caching.Secondly,it aims at the comprehensive performance and the underlying computing resources of the system.Finally,a parallel power flow algorithm for distributed computing is proposed.The parallel update of the Jacobian array and the parallel iteration of the modified equation are implemented in Spark cloud.(1)By analyzing the source code and introducing two characteristic parameters of the Spark operation stream,the dynamic priority screening of the task is realized;Based on the analysis and optimization of the task structure,combined with the distributed characteristics of the RDD,improving the operational efficiency of the task under limited resources.(2)By analyzing the communication mechanism between RDD nodes,the hierarchical scheduling strategy of Spark computing flow is established to achieve high performance computing,cost reduction and load balancing.A multi-objective optimization algorithm considering preference regions is proposed.Simulation tests shows that the overall energy efficiency of the algorithm is better than traditional algorithms such.(3)By Including: correction of the correction amount related to the sparse matrix,and Distributed multiplication of high dimensional matrices.Finally,a parallel computing cluster consisting of Spark and Hadoop is implemented.The feasibility and effectiveness of the algorithm are verified in IEEE synthesis system.