[Paper Review] TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

Haixu Wu, Tengge Hu+4 more|arXiv (Cornell University)|Oct 5, 2022

Time Series Analysis and Forecasting420 citations

TL;DR

TimesNet converts 1D time series into multiple 2D tensors based on learned periods to model intraperiod and interperiod variations with 2D kernels, achieving state-of-the-art results across forecasting, imputation, classification, and anomaly detection. It introduces TimesBlock as a modular, task-general backbone that adaptively learns multi-periodicity.

ABSTRACT

Time series analysis is of immense importance in extensive applications, such as weather forecasting, anomaly detection, and action recognition. This paper focuses on temporal variation modeling, which is the common key problem of extensive analysis tasks. Previous methods attempt to accomplish this directly from the 1D time series, which is extremely challenging due to the intricate temporal patterns. Based on the observation of multi-periodicity in time series, we ravel out the complex temporal variations into the multiple intraperiod- and interperiod-variations. To tackle the limitations of 1D time series in representation capability, we extend the analysis of temporal variations into the 2D space by transforming the 1D time series into a set of 2D tensors based on multiple periods. This transformation can embed the intraperiod- and interperiod-variations into the columns and rows of the 2D tensors respectively, making the 2D-variations to be easily modeled by 2D kernels. Technically, we propose the TimesNet with TimesBlock as a task-general backbone for time series analysis. TimesBlock can discover the multi-periodicity adaptively and extract the complex temporal variations from transformed 2D tensors by a parameter-efficient inception block. Our proposed TimesNet achieves consistent state-of-the-art in five mainstream time series analysis tasks, including short- and long-term forecasting, imputation, classification, and anomaly detection. Code is available at this repository: https://github.com/thuml/TimesNet.

Motivation & Objective

• Motivate modeling of temporal variations through multi-periodicity in time series.
• Propose a 2D representation of time series to unify intraperiod- and interperiod-variations.
• Introduce TimesNet and TimesBlock as a modular, task-general backbone.
• Show state-of-the-art performance across forecasting, imputation, classification, and anomaly detection.

Proposed method

• Estimate multiple periods via FFT-based frequency amplitude analysis to identify top frequencies and corresponding periods.
• Transform the 1D time series into multiple 2D tensors using reshape with identified period lengths, enabling 2D variation modeling.
• Process each 2D tensor with a shared inception-style block to capture multi-scale intraperiod and interperiod variations in 2D space.
• Aggregate outputs from different period-based 2D views using a softmax-weighted fusion inspired by amplitude-derived importance.
• Stack TimesBlocks in a residual fashion to progressively refine representations.

Experimental results

Research questions

• RQ1 Can multi-periodicity be leveraged to decompose complex temporal variations into intraperiod and interperiod components?
• RQ2 Does transforming 1D time series into 2D tensors and applying 2D kernels improve representation learning for diverse TS tasks?
• RQ3 Is TimesNet a task-general backbone that achieves state-of-the-art across forecasting, imputation, classification, and anomaly detection?
• RQ4 How does adaptive aggregation based on period amplitudes affect performance across tasks?

Key findings

• TimesNet achieves consistent state-of-the-art across five mainstream time series analysis tasks.
• A modular TimesBlock can discover multi-periodicity and extract temporal 2D-variations via a parameter-efficient inception block.
• Transforming 1D series to 2D tensors enables effective modeling of intraperiod- and interperiod-variations with 2D kernels.
• Replacing the inception block with stronger 2D backbones further improves performance, validating the 2D-variation design.
• TimesNet demonstrates task-generality as a foundation model for time series analysis.