Parallel Processing And Diagnosis Of Big Data In Substation Equipment Monitoring Based On Storm

Transformer equipment is one of the most important equipment in the power system.Whether its fault can be diagnosed quickly and accurately will directly affect the reliability of network power supply and the safe operation of the system.Continuous online power grid monitoring data has challenged traditional data processing methods,and the power system has entered the "big data" era.Storm,as a distributed processing framework for stream data,has the advantages of real-time and scalability,which can meet the real-time and continuous monitoring needs of grid data.This paper studies the parallel decomposition method of partial discharge signals based on Storm,improves the online sequence learning machine algorithm,and studies the application of model markup language and memory virtual distributed storage system Alluxio across platforms.This paper designs a parallel signal processing method based on Storm’s Ensemble Empirical Mode Decomposition(EEMD),and implements two different parallel discharge signal processing methods of waveform segmentation parallel and EMD process parallel.The performance and application scenarios of the two EEMD parallel methods are compared through experiments.At the same time,the performance bottleneck of the Storm cluster is noticed,and the flow control strategy of the data flow is proposed to avoid thread blocking and maximize the cluster processing performance.Using the advantages of Storm distributed real-time computing,the problem of large complexity of the algorithm itself is well solved,and the real-time nature of online processing can be satisfied.Online sequence extreme learning machine(OS-ELM)algorithm is deeply studied.Based on the analysis of its parallelism,combined with the Storm framework to complete the parallel learning process,the model learning speed can be improved to a certain extent.On the basis of parallelization,using the idea of ensemble learning,different models with different activation functions are used for learning and diagnosis,and error weights and gradient descent algorithms are introduced to adjust the output of different models to improve the accuracy of prediction.The model was verified using DGA data.The results show that the model can meet the needs of fault diagnosis.The accuracy of classification and the training speed of the model are better than the traditional ELM algorithm.