Research On Transformer Fault Diagnosis Based On Spark Technology

Traction transformers are widely used in railway substations to realize the transition of grid voltage to traction power supply system voltage.It is an important equipment for safe and stable operation of traction power supply system.In the traction power supply system,using online monitoring technology can obtain the operation data of transformer,understand the operation status of transformer,diagnose potential faults in time and effectively,and reduce the probability of failure of the traction power supply system.The online monitoring technology contains multiple monitoring points,which can produce many monitoring data in a short time.The volume of data is diversified in terms of volume,velocity,variety,value,veracity.Traditional data analysis and processing methods can not meet the requirements of processing large amounts of data in time and space.Parallel processing of large data analysis methods can achieve efficient analysis of massive data.Different from the traditional small batch data processing,Spark is an open source distributed big data computing framework.Its calculation is fast,based on memory,supports many programming languages,etc.It can calculate massive data at high speed,providing a new method to solve the problem of processing massive data in stand-alone mode.The paper firstly introduces the common methods of transformer fault diagnosis and big data technology.and puts forward the traction transformer fault diagnosis method based on Spark according to the advantages and disadvantages of each theory.Secondly,based on the analysis and research of transformer fault diagnosis at home and abroad,the SVM transformer fault diagnosis method based on particle swarm optimization is adopted.Then,the paper designs and implements the parallelization process of PSO-SVM,designs the real-time update of MSSQL database information in Kafka message queue,forms the real-time input of Spark Streaming data,and designs the transformer fault diagnosis platform based on Spark.Finally,the platform uses dissolved gas data from transformer oil to verify the experiment,and discusses the experimental results from the aspects of accuracy,acceleration ratio and running time.The results show that with the increase of the amount of data,in the case of ensuring the accuracy of the stability,the parallel algorithm has a faster processing speed than the stand-alone mode and the accuracy of the algorithm is equivalent,which proves that the Spark platform based on PSO-SVM has a greater advantage in processing massive data.