Parallel Implementation Of Partial Discharge Pattern Recognition For Electrical Equipment Based On Spark Framework

Partial discharge is closely linked to the state of the insulation of electrical equipment.The causes and discharge positions of different types of partial discharges are not the same,and the degree of damage to the equipment is also different.Therefore,efficient identification of partial discharge types is of great significance for electrical equipment maintenance personnel to determine the discharge location and to plan maintenance tasks.Based on the analysis of partial discharge characteristics,this paper mainly studies the feature extraction and classification methods of partial discharge signals of power equipment,and uses distributed technology to realize parallel algorithm to process data.The main work contents are as follows:A feature extraction method based on ensemble empirical mode decomposition and multi-scale sample entropy is proposed.The four partial discharge signals are processed by the ensemble empirical mode decomposition algorithm,and the intrinsic mode functions in different frequency bands are decomposed,and the five intrinsic mode functions after dimensionality reduction are obtained.Then,the corresponding multi-scale sample entropy is obtained for the intrinsic mode functions,and the feature vectors of the original data are combined and extracted.Finally,the feature vector is input into the support vector machine to implement classification.The experimental results show that the features extracted by this method can effectively represent the partial discharge signal,and have a high recognition rate and strong stability.Support vector machine(SVM)algorithm is deeply studied.For large data samples of power grid,a parallel framework based on Spark is proposed for classification and recognition.Firstly,a Cascade-based vector machine model is proposed.By dividing data sets and using the idea of dividing and conquering,the sub-support vector machine model is trained separately.Then,the final classification machine model is assembled to complete the parallelization of the whole classification and recognition and improve the processing speed under large data.Since the traditional support vector machine is based on two classifications,the partial discharge signal is a sample of multiple categories.Therefore,the multi-classification problem of support vector machine is studied,and the multi-classification model based on decision tree is discussed in detail.The Spark is designed to realize the multi-classification support vector machine model in one-to-one and one-to-many mode,and the standard data set is used to compare thesingle machine and the multi-machine to realize parallel processing of multiple classifications and improve the processing rate.Finally,the model is applied to the classification of partial discharge classification.