Hybrid Neural Networks over Time Series for Trend Forecasting

Tao Lin, Tian Guo, Karl Aberer

Abstract:

The trend of time series characterize the intermediate upward and downward patterns of time series. Learning and forecasting the trend in time series data play an important role in many real applications, ranging from resource allocation in data centers, load schedule in smart grid and so on. Inspired by the recent successes of neural networks, in this paper we propose TreNet, a novel hybrid neural network based learning approach over time series and the associated trend sequence. TreNet leverages convolutional neural networks (CNNs) to extract salient features from local raw data of time series and uses a long-short term memory recurrent neural network (LSTM) to capture the sequential dependency in historical trend evolution. Some preliminary experimental results demonstrate the advantage of TreNet over cascade of CNN and LSTM, CNN, LSTM, Hidden Markov Model method and various kernel based
baselines on real datasets.

Conflicts: epfl.ch