Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks

Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks

July 2018, Ann Arbor, MI, USA | Guokun Lai, Wei-Cheng Chang, Yiming Yang, Hanxiao Liu
This paper proposes a novel deep learning framework called LSTNet for multivariate time series forecasting, which effectively captures both long-term and short-term temporal patterns. LSTNet combines Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) to extract short-term dependencies and long-term trends. It also integrates a traditional autoregressive model to address scale insensitivity. The framework is evaluated on real-world datasets with complex patterns, achieving significant performance improvements over existing methods. LSTNet uses a convolutional layer to detect local dependencies and a recurrent layer to capture long-term dependencies. A novel recurrent-skip structure is introduced to handle very long-term dependencies. Additionally, a temporal attention layer is used to enhance the model's ability to focus on relevant patterns. The autoregressive component ensures robustness to scale changes in input signals. The model outperforms traditional linear models and GRU-based networks in forecasting tasks. The framework is effective in capturing both short-term and long-term patterns, making it suitable for applications such as traffic forecasting, solar energy prediction, and electricity consumption forecasting. The study demonstrates that LSTNet is robust and efficient in handling complex time series data with mixed patterns.This paper proposes a novel deep learning framework called LSTNet for multivariate time series forecasting, which effectively captures both long-term and short-term temporal patterns. LSTNet combines Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) to extract short-term dependencies and long-term trends. It also integrates a traditional autoregressive model to address scale insensitivity. The framework is evaluated on real-world datasets with complex patterns, achieving significant performance improvements over existing methods. LSTNet uses a convolutional layer to detect local dependencies and a recurrent layer to capture long-term dependencies. A novel recurrent-skip structure is introduced to handle very long-term dependencies. Additionally, a temporal attention layer is used to enhance the model's ability to focus on relevant patterns. The autoregressive component ensures robustness to scale changes in input signals. The model outperforms traditional linear models and GRU-based networks in forecasting tasks. The framework is effective in capturing both short-term and long-term patterns, making it suitable for applications such as traffic forecasting, solar energy prediction, and electricity consumption forecasting. The study demonstrates that LSTNet is robust and efficient in handling complex time series data with mixed patterns.
Reach us at info@study.space
Understanding Modeling Long- and Short-Term Temporal Patterns with Deep Neural Networks