Respecting Time Series Properties Makes Deep Time Series Forecasting Perfect

TL;DR

This paper emphasizes the importance of respecting time series properties in deep learning models, introduces RTNet, and demonstrates its superior performance and efficiency across multiple benchmarks.

Contribution

It provides a rigorous analysis of common deep time series forecasting methods and proposes RTNet, a novel network that respects time series properties for improved accuracy and efficiency.

Findings

RTNet outperforms state-of-the-art baselines in accuracy.

RTNet is more efficient in time complexity and memory usage.

Respecting time series properties enhances forecasting stability and performance.

Abstract

How to handle time features shall be the core question of any time series forecasting model. Ironically, it is often ignored or misunderstood by deep-learning based models, even those baselines which are state-of-the-art. This behavior makes their inefficient, untenable and unstable. In this paper, we rigorously analyze three prevalent but deficient/unfounded deep time series forecasting mechanisms or methods from the view of time series properties, including normalization methods, multivariate forecasting and input sequence length. Corresponding corollaries and solutions are given on both empirical and theoretical basis. We thereby propose a novel time series forecasting network, i.e. RTNet, on the basis of aforementioned analysis. It is general enough to be combined with both supervised and self-supervised forecasting format. Thanks to the core idea of respecting time series properties, no matter in which forecasting format, RTNet shows obviously superior forecasting performances compared with dozens of other SOTA time series forecasting baselines in three real-world benchmark datasets. By and large, it even occupies less time complexity and memory usage while acquiring better forecasting accuracy. The source code is available at https://github.com/OrigamiSL/RTNet.