Abstract:Time series, characterized by a sequence of data points arranged in a discrete-time order, are ubiquitous in real-world applications. Different from other modalities, time series present unique challenges due to their complex and dynamic nature, including the entanglement of nonlinear patterns and time-variant trends. Analyzing time series data is of great significance in real-world scenarios and has been widely studied over centuries. Recent years have witnessed remarkable breakthroughs in the time series community, with techniques shifting from traditional statistical methods to advanced deep learning models. In this paper, we delve into the design of deep time series models across various analysis tasks and review the existing literature from two perspectives: basic modules and model architectures. Further, we develop and release Time Series Library (TSLib) as a fair benchmark of deep time series models for diverse analysis tasks, which implements 24 mainstream models, covers 30 datasets from different domains, and supports five prevalent analysis tasks. Based on TSLib, we thoroughly evaluate 12 advanced deep time series models on different tasks. Empirical results indicate that models with specific structures are well-suited for distinct analytical tasks, which offers insights for research and adoption of deep time series models. Code is available at <a class="link-external link-https" href="https://github.com/thuml/Time-Series-Library" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

This paper focuses on a comprehensive investigation and benchmark testing of deep time series models. With the extensive use of time series data in real-world applications such as financial risk assessment, energy sustainability, and weather forecasting, analyzing these data has become increasingly important. Traditional time series analysis methods, such as ARIMA and exponential smoothing, have limitations in handling complex nonlinear relationships and long-term dependencies, although they play a role in discovering patterns and trends. In recent years, deep learning models, especially Transformer models with attention mechanisms, have shown excellent performance in capturing complex dependencies in time series data. However, there is currently a lack of a comprehensive time series model library for fair performance evaluation across different tasks. Therefore, the paper proposes the Time Series Library (TSLib), which implements 24 mainstream models, covering 30 different domains of datasets and supporting 5 common analysis tasks. Through experimental evaluations on TSLib, the authors found that models designed for specific analysis tasks perform better, providing guidance for research and application of deep time series models. The main contributions of the paper include: 1. A comprehensive investigation of the design of deep time series models, covering basic modules and model architectures. 2. Development and release of TSLib, a fair benchmarking library for deep time series models. 3. Experimental results based on TSLib reveal performance differences of different models on different tasks, providing a basis for selecting appropriate models. Through this work, researchers and practitioners can better understand how to leverage deep learning techniques to tackle challenges in time series data analysis.

Deep Time Series Models: A Comprehensive Survey and Benchmark

Deep Time Series Forecasting Models: A Comprehensive Survey

Deep Time Series Models for Scarce Data

Large Models for Time Series and Spatio-Temporal Data: A Survey and Outlook

Time Series Dataset Survey for Forecasting with Deep Learning

Foundation Models for Time Series Analysis: A Tutorial and Survey

A Review of Deep Learning Models for Time Series Prediction

A Survey on Deep Learning based Time Series Analysis with Frequency Transformation

Learning Deep Time-index Models for Time Series Forecasting

A Survey of Time Series Foundation Models: Generalizing Time Series Representation with Large Language Model

A Survey on Time-Series Pre-Trained Models

MvTS-library: an Open Library for Deep Multivariate Time Series Forecasting

A Survey of Deep Learning and Foundation Models for Time Series Forecasting

PyDTS: A Python Toolkit for Deep Learning Time Series Modelling

Effective LSTMs with Seasonal-Trend Decomposition and Adaptive Learning and Niching-Based Backtracking Search Algorithm for Time Series Forecasting

A Comprehensive Survey of Time Series Forecasting: Architectural Diversity and Open Challenges

TSCF: An Improved Deep Forest Model for Time Series Classification

FoundTS: Comprehensive and Unified Benchmarking of Foundation Models for Time Series Forecasting

Universal Time-Series Representation Learning: A Survey

Deep Unsupervised Domain Adaptation for Time Series Classification: a Benchmark