Respecting Time Series Properties Makes Deep Time Series Forecasting Perfect
This addresses inefficiencies in deep learning-based time series forecasting for applications requiring accurate predictions, though it appears incremental as it builds on existing methods by focusing on overlooked properties.
The paper tackles the problem of deep time series forecasting models ignoring time series properties, leading to inefficiency and instability, and proposes RTNet, which shows superior performance with better accuracy, lower time complexity, and memory usage compared to dozens of SOTA baselines on three real-world datasets.
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.