多變量卷積類神經網路於時間序列資料分類

Abstract

卷積類神經網路在圖像辨識方面有強大的效能和進步。卷積類神經網路 可以從資料中萃取出特徵，因此在許多資料辨識中能有很好的表現。但卷積類神經網路在時間序列資料上的應用還存有許多挑戰，其中一個問題是圖像辨識和時間序列資料分類資料特徵就有所不同。我們提出一個資料轉變得版模，將時間序列資料轉換成三維資料，另外還提出一個多變量卷積類神經網路用於時間序列資料分類問題。我們此神經網路和其他優異的演算法做比較，在PHM 2015 challenge 資料集上，多變量卷積類神經網路在分數和預測的精準率上表現亮眼。另外我們還將此架構應用在occupancy detection 資料集上，其準確度也優於Random forest。

In recent years, the convolutional neural network (CNN) has achieved great success on image classification, since it learns feature representation from a collection of images. Recent studies have shown that machine learning algorithms could benefit from good feature representation, explaining why CNN, which considers feature learning and classification in the same network, has achieved breakthrough performance in many pattern recognition tasks. However, it is still a challenge to apply CNN to time series data. One of the problems is that the time series data and image data possess different characteristics. This thesis proposes a tensor scheme and a novel deep learning architecture called multivariate convolutional neural network for time series classification. We evaluate our proposed method with PHM 2015 challenge data, and compare with several state-of-the-art algorithms, including gradient boost tree, random forest and logistic regression. The experimental results indicate that the proposed method outperforms the other alternatives on the scores and false positives. dditionally, the proposed method achieves better accuracy than random forest on occupancy detection data.