Long-term Time Series Forecasting based on Decomposition and Neural Ordinary Differential Equations

Seonkyu Lim,Jaehyeon Park,Seojin Kim,Hyowon Wi,Haksoo Lim,Jinsung Jeon,Jeongwhan Choi,Noseong Park

2023-11-10

Abstract:Long-term time series forecasting (LTSF) is a challenging task that has been investigated in various domains such as finance investment, health care, traffic, and weather forecasting. In recent years, Linear-based LTSF models showed better performance, pointing out the problem of Transformer-based approaches causing temporal information loss. However, Linear-based approach has also limitations that the model is too simple to comprehensively exploit the characteristics of the dataset. To solve these limitations, we propose LTSF-DNODE, which applies a model based on linear ordinary differential equations (ODEs) and a time series decomposition method according to data statistical characteristics. We show that LTSF-DNODE outperforms the baselines on various real-world datasets. In addition, for each dataset, we explore the impacts of regularization in the neural ordinary differential equation (NODE) framework.

Machine Learning

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address the challenging issue in **Long-term Time Series Forecasting (LTSF)**. Specifically, the paper proposes a new model named **LTSF-DNODE** to overcome the limitations of existing methods and improve forecasting performance.

#### Limitations of Existing Methods:
1. **Transformer-based methods**: Although they perform well in certain tasks, they suffer from the problem of losing temporal information, especially in long sequence forecasting.
2. **Linear methods**: Despite showing excellent performance in recent studies, these models are too simplistic to fully exploit the complex characteristics of datasets.

#### Proposed Method:
LTSF-DNODE combines methods based on linear ordinary differential equations (ODE) and time series decomposition techniques. Through this approach, the model can better capture the dynamic characteristics in the data and demonstrate superior performance on various real-world datasets.

#### Main Contributions:
1. **Time Series Decomposition**: Analyzing the temporal information of each real-world dataset, detecting seasonality, and performing appropriate decomposition accordingly.
2. **Neural Ordinary Differential Equation Framework**: Utilizing the NODE framework to transform discrete linear layers into continuous modeling, thereby providing better regularization capabilities.
3. **Experimental Proof**: Demonstrating the effectiveness of the NODE framework and decomposition based on data characteristics through extensive experiments.

In summary, LTSF-DNODE aims to improve the accuracy of long-term time series forecasting by introducing more sophisticated data processing methods, particularly when dealing with datasets with complex characteristics.