Scale-Aware Neural Architecture Search for Multivariate Time Series Forecasting

• DOI: 10.48550/arxiv.2112.07459
• Publication Date: 2021-01-01
• Authors: Donghui Chen; Ling Chen; Zongjiang Shang; Youdong Zhang; Bo Wen; Yang Cheng-hu
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Machine Learning (cs.LG)

Multivariate time series (MTS) forecasting has attracted much attention in many intelligent applications. It is not a trivial task, as we need to consider both intra-variable dependencies and inter-variable dependencies. However, existing works are designed for specific scenarios, require domain knowledge expert efforts, which difficult transfer between different scenarios. In this paper, propose scale-aware neural architecture search framework MTS (SNAS4MTF). A multi-scale decomposition module transforms raw into sub-series, can preserve temporal patterns. An adaptive graph learning infers the under scales without any prior knowledge. For forecasting, space capture at each scale. The decomposition, learning, modules jointly learned an end-to-end framework. Extensive experiments on two real-world datasets demonstrate that SNAS4MTF achieves promising performance compared with state-of-the-art methods.