| Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 107 === Time series classification (TSC) has drawn a lot of attention in different domains such as finance and health informatics. This study considers the application of deep learning for the classification of abnormal patterns in time series of product quality measurements to assist in the diagnosis of quality problems. Specifically, we propose a novel Multi-Channel Deep Convolutional Neural Networks (MC-DCNN) framework for time series classification. The input vectors of each channel include the raw time series data as well as the transformed data as alternative representations of raw data. The idea is to obtain diversified features from different representations of raw time series data. In this study, we propose using statistical method and wavelet multiresolution analysis as data representation methods.
The proposed model first learns features from individual univariate time series in each channel, and combines information from all channels as feature representation at the final layer. Then, the learnt features are fed into a Multilayer Perceptron (MLP) for classification. In this study, we also explore the methods of encoding time series as 2D texture images to discover different feature types that are not available for 1D signals. The methods we consider include Gramian Angular Summation/Difference Fields (GASF/GADF), Markov Transition Fields (MTF) and recurrence plot.
The proposed method was evaluated using data collected from industry. The results indicate that the proposed method can significantly improve the classification accuracy.
|
|---|
