Centralized Parallel Processing And Diagnosis Of Power Equipment Monitoring Data Based On Cloud Computing

Data integration and information sharing are the inevitable trend of building a strong smart grid.With the dispatching center of a power grid being evolved towards to the integration of a power dispatch center and a remote control center,more and more condition monitoring data of the apparatuses need to be sent to the center.Fast processing of large-scale historical and real-time online condition monitoring data are being challenged,therefore this dissertation focuses on the problem of centralized parallel processing and diagnosis of power equipment condition data based on cloud computing and big data technology.A multi-application scenarios oriented integrated cloud computing platform architecture is designed for data processing of power equipment condition monitoring.In this platform,a variety of parallel computing frameworks share the same set of IT infrastructure,which not only saves the computing platform investment and maintenance costs,but also facilitates data integration and information sharing.All application tasks are assigned to the most appropriate framework with the specified resources,based on processing mode of data and real-time demands of users.An adaptive algorithm to extract the basic parameters for partial discharge signals is proposed.On this basis,this dissertation proposes a centralized parallel batch method for huge mass of partial discharge signals based on Hadoop Map Reduce.The local extreme points of partial discharge signals are filtered by an amplitude threshold and another interval threshold,which are both adaptive.By this way,three parameters of partial discharge,namely frequency,quantity and phase,are extracted automatically.By means of Map Reduce framework,the whole process of analyzing partial discharge signals,including basic parameters extraction,pattern depiction,features calculation and discharge type recognition,is parallelized,improving the processing efficiency of large volume data.A parallelized version of ensemble empirical mode decomposition(EEMD)is proposed based on Spark in-memory computing technology,being an important supplement to Hadoop Map Reduce which cannot be applied to complex data processing scenarios.The parallelism of an EEMD procedure is analyzed,and two different structure of parallel EEMD algorithm,namely epoch parallel and trial parallel,are designed and implemented.Then the proposed parallel algorithm is applied to feature extraction of partial discharge waveform signals,and compared with the serial EEMD algorithm and the parallel EEMD algorithm based on Hadoop Map Reduce in terms of computational performance.An online parallel fault diagnosis method for power equipment is proposed based on Storm real-time processing technology and variable predictive model based class discriminate(VPMCD).In order to meet the needs of online diagnosis,the incremental learning mechanism is introduced into VPMCD method,and the incremental update of the variable predictive models(VPMs)is implemented by the recursive least-squares algorithm.Taking into account the problem of online fault diagnosis for a large number of power equipments,a real-time processing framework for large streams of high velocity data is designed based on Storm,then the initialization,incremental learning and application of VPMs is achieved performing on Storm by constructing the monitoring mechanism on components of Storm topology.Taking dissolved gas analysis of transformer as an application,the classification and computational performance of the proposed method are tested.