Parallel Diagnosing Method For Monitoring Big Data Of Electric Power Equipment Based On Spark Framework

With the development of smart grids,intelligent detection of power equipment faults has become a research hotspot.All kinds of intelligent monitoring equipment are increasing day by day.Once the power equipment fails,a large amount of fault data will be generated.Especially in extreme weather,blowout data will be generated.The traditional single machine processing environment can not longer cope.In order to solve this problem,this paper combines traditional fault diagnosis methods with Spark and Map Reduce big data frameworks,and uses transformer fault diagnosis as an example to study the power equipment diagnosis methods in the big data environment.This paper proposes a parallel algorithm of deep belief network on Spark framework and uses it for transformer fault diagnosis.The parallelization of the DBN model is mainly deployed from four aspects: distributed storage of data on RDD,parallelized structure,data migration,and tuning of broadcast variable mechanism.At the same time,based on the transformer oil chromatographic analysis method,the performance of the DBN,the parallel DBN under the Spark framework and the DBN under the Map Reduce framework are compared and analyzed experimentally,which proves the effectiveness of the parallel DBN algorithm under the Spark framework.A transformer fault diagnosis method based on genetic algorithm optimized deep belief network(GA-DBN)is proposed.In order to further improve the accuracy of transformer fault diagnosis and solve the problem that the deep belief network(DBN)easily falls into local optimum,a GA-DBN model is proposed.Because the genetic algorithm has the characteristics of global optimization,the initial parameter value is found by the genetic algorithm,then the DBN model is trained in the local solution space,and finally the network is fine-tuned by the BP algorithm.Combining two methods to diagnose power transformer faults,the case analysis shows that the method has faster convergence speed and higher diagnostic accuracy.A method of information sharing between cloud platforms is designed.The model learned by Spark is exported as a file in PMML format,which is prepared for the real-time analysis of the Storm framework.