Research On Memory Cluster Computing Of Electric Power Cloud Platform

With the development of smart grid and energy Internet, the scale of power systems continues to expand, electric power data rapidly grow with an astonishing speed and the diversified structure, these complex data almostly come from smart metering, distribution automation, digital protection devices and data relating to smart grid implementation process. Faced with such a large and diverse data, cloud computing and big data processing technology provides a whole new set of technical means for electric power data analysis, and electric power cloud platform is getting hot, but it does not meet the needs of high computing performance of power system.This article analyzes the advantages and disadvantages of several big data processing technologies based on the background of electric power big data.Currently, Hadoop is the hotspot of research on big data, and has a higher efficiency on batch data processing. However, there are a lot of data latency and read/write latency when we process data with Map Reduce technology, so it is difficult to meet the needs of high computing performance. Spark is a clustering technology based on memory, and suitable for intensive computation and iterative computation. Calculation speed is fast. In connection with high computing performance demands for electric power system, this article designs a framework for distributed memory computing of power system with introducing memory cluster technology. The data, which is cllected by intelligent electronic devices, intelligent electric meter and phasor measurement units, is stored in distributed file system, and analyzed by the memory frame Spark. And we introduce distributed memory file system to the framework in order to reduce the time of access disk.This paper analyzes the characteristics of power flow calculation classic algorithm and intelligent algorithm. In consideration of the application of the universality, convergence and stability, we use Newton-Raphson method for power flow calculation. This article presentes a power flow analysis optimization method, which is based on distributed data collection(RDD) and a directed acyclic graph(DAG), through the study of Newton-Raphson power flow. The data generated during power flow calculation process is saved in the RDD, so we can reads data in the RDD by the speed of reading memory when we use the method, which reduces read/write time. Meanwhile, the power flow calculation procedure is optimized by DAG, so narrow-dependent steps are excuted in the same node, and wide-dependent steps are executed in parallel. Then we can effectively reduce data transmission time and running time in the process of poer flow calculation.Lastly, we built experimental environment of Spark On Yarn, implement Newton method power flow calculation in Java programming language based on Map Reduce technology. Also, we realize the power flow calculation in the mode of Spark clusters(NBRD method) and stand-alone by the Scala programming language. This paper constructs and simulates a large-scale data set based on the standard test data of IEEE, and conducts the experiment in many ways: the scale of cases, the speed ratio of power flow calculation time, the scale of Spark cluster. It is showed by the test results that the times of itteration is stable, the result is correct, and when the scale of examples and the cluster is small, the advantages of this method are not obvious. But computing time decreases and speed ratio increases when the scale of examples cluster expands, which indicates that the performance is superior to stand-alone mode and Map Reduce mode.