Research On Key Issues In Massive Information Processing Of Smart Grid

A large amount of multi-source, heterogeneous and redundant data is generated withthe rapid development of smart grid. It is a tough task for present control andprocessing system to utlized these massive data. As a result, abnormal state or faultscan not be discovered. In this paper, we aimed at key technologies in massive dataprocessing of smart grid. Feasibility of adapting Hadoop to smart grid is analyzedfirstly. We take fault recorder file as an example to illustrate how files are sampled andsaved in distributed computing system. With the distributed storage platform, theexisting processing technologies of massive data, such as data stream, sliding windowtechnology, data mining and MapReduce, are adopted to our system.In this paper, we proposed a method named adaptive sliding window to detect powerquality disturbance whicn is typical data stream. Wavelet transform is used to denoising,detecting disturbance and extracting feature vector. Sliding window found bycomprehensive detection method can adjust its length to cover the entire disturbance.Hoeffding Tree is a typical data mining method which is demonstrated to be effectivein classification of disturbance. Power system with wind plants has special feature inprotection system. Failures in protection can be caused by wind plant capacity, windspeed, crowbar, fault location, transition resistance and phase angle which is a typicalassociation rule problem. In this paper, concept hierarchy is introduced to preprocessthe attributes. Mining algorithm of multi-layered and multi-dimensional associationrule is adopted to analyze the relationship between influences and results above.Fault information is one of the most typical data in smart grid. It is proposed that a faultinformation system based on Hadoop can meet the demands of processing massive data.This system is composed of a distributed data acquisition module, a cloud computingmodule and a decision-making module. It is efficient for this system to save andretrieve fault information. It is also implemented in this system that original data arepreprocessed in parallel and faults are diagnosed.Finally, every scheme introduced above is demonstrated by a simulation. Fault recorderfiles are continuously generated by RTDS and saved in distributed file system based on Hadoop which is easy to retrieve. Power quality disturbance models are given togenerated data stream which can be processed by adaptive sliding window andHoeffiding Tree algorithm. Power system with wind plants are simulated andassociation rules are mined. Typical four-station system are built by RTDS anddifferent recorder files are saved and analyzed to comprehensive fault diagnosis.