Multivariate Time Series Data Transformation for Convolutional Neural Network

• Main Authors: Chen-Yi YANG, 楊晨宜
• Other Authors: Chao-Lung Yang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/sd989v

碩士 === 國立臺灣科技大學 === 工業管理系 === 106 === This thesis proposes a novel framework to encode time series data into two-dimensional images, and aggregate the images into one image for each batch of data. After transformation and aggregation, the images passed through a convolutional neural network, which is outstanding in dealing with computer vision problems. The convolutional neural network learned the image features and predict the status. This study applied four methods to encode time series into images: Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP). The overlaying and appending methods to aggregate the images were validated and evaluated using open datasets. The results of the experiments find that encoding time series data into images and aggregating the images by the appending method are helpful in increasing prediction accuracy.