Improving the Transferability of Time Series Forecasting with Decomposition Adaptation

TL;DR

This paper introduces SeDAN, a transfer learning architecture that decomposes time series features into seasonal and trend components to improve forecasting performance across domains, especially with limited data.

Contribution

The paper proposes a novel transfer architecture, SeDAN, with implicit contrastive decomposition and domain-specific adaptation methods for better transferability in time series forecasting.

Findings

SeDAN outperforms existing methods on five benchmark datasets.

Decomposition into seasonal and trend features enhances transferability.

The approach provides more stable and efficient knowledge transfer.

Abstract

Due to effective pattern mining and feature representation, neural forecasting models based on deep learning have achieved great progress. The premise of effective learning is to collect sufficient data. However, in time series forecasting, it is difficult to obtain enough data, which limits the performance of neural forecasting models. To alleviate the data scarcity limitation, we design Sequence Decomposition Adaptation Network (SeDAN) which is a novel transfer architecture to improve forecasting performance on the target domain by aligning transferable knowledge from cross-domain datasets. Rethinking the transferability of features in time series data, we propose Implicit Contrastive Decomposition to decompose the original features into components including seasonal and trend features, which are easier to transfer. Then we design the corresponding adaptation methods for decomposed features in different domains. Specifically, for seasonal features, we perform joint distribution adaptation and for trend features, we design an Optimal Local Adaptation. We conduct extensive experiments on five benchmark datasets for multivariate time series forecasting. The results demonstrate the effectiveness of our SeDAN. It can provide more efficient and stable knowledge transfer.