Open AccessTransformer Fault Diagnosis Based on Optimized CPSO-BP Neural Network
Author(s) -
Jiyan Zou,
Quan Liang,
Xiaoming Xu,
Qiang Wang
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/781/4/042047
Subject(s) - artificial neural network , computer science , cluster analysis , gradient descent , particle swarm optimization , transformer , algorithm , fault (geology) , convergence (economics) , backpropagation , artificial intelligence , data mining , engineering , voltage , seismology , geology , economic growth , electrical engineering , economics
Oil chromatographic analysis (DGA) is an important way to transformer fault diagnosis, combining research topics based on I-K-means clustering, t SNE visual clustering in data mining, fault classification number. In order to improve the convergence speed of neural network, an improved back-propagation BP neural network using ADAM gradient optimization algorithm instead of traditional stochastic gradient descent optimization to update the weight of neural network is proposed. Fuzzy C-means clustering and particle swarm optimization are proposed to optimize the initial parameters of neural network. By using 3500 data samples of transformers from a power plant in a city of Liaoning Province to carry out simulation experiments, and comparing the traditional BP network algorithm, CPSO-BP network algorithm and the CPSO-BP network algorithm optimized by Adam, it is proved that the CPSO-BP network optimized by Adam has fast training convergence, strong generalization ability and high accuracy. At the same time, the accuracy, precision, recall and F-1 values were used to evaluate the CPSO-BP network algorithm optimized by ADAM to verify the effectiveness and stability of the algorithm in transformer fault diagnosis.