Earticle

초록

영어

This paper presented a fault detection method based on deep learning Convolutional Neural Networks(CNN) and Long Short-Term Memory. Using CNN we get more abstract features representation in the higher level to find the distributed characteristics of the data. After obtaining the features, use LSTM to further mining useful information in the time dimension. First, we presented a CNN model which has 9 layers to extract more abstract features. By comparing three different CNN models, we realized that the shape of the original data set is much important. 16×16 shape of data set has high accuracy, it is 95%. Also comparing with traditional fault detection model, it is much better than random forest and Deep Neutral network(DNN). And the results show that the proposed CNN model can extract the features automatically for fault detection intelligently. However, data has a complex time correlation with each other. How to get the most information in the data for fault detection? We presented LSTM to extract more useful information in the time dimension. The proposed CNN-LSTM method has the highest accuracy which up to 96.13%. The proposed CNN-LSTM exhibits the best performance in the electric vehicle charging pile diagnosis.

목차

Abstract
1. Introduction
2. Relevant Theoretical
2.1 Convolution neural network
2.2 Long Short-Term Memory
3. Experiment
3.2 Implemented Convolution Neural Network
3.3 Implemented Proposed CNN-LSTM
4. Conclusions
References

키워드

저자

• Xiaorui Shao [ Department of IT Convergence and Application Engineering, Pukyong National University, Busan, South Korea ]
• Chang-Soo Kim [ Department of IT Convergence and Application Engineering, Pukyong National University, Busan, South Korea ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  IJICTDC [International Journal of Information Communication Technology and Digital Convergence]

• 간기

  반년간

• pISSN

  2466-0094

• 수록기간

  2016~2026

• 십진분류

  KDC 300 DDC 303