Two-Stage Framework for Seasonal Time Series Forecasting

TL;DR

This paper introduces a two-stage neural network framework for seasonal time series forecasting that captures long-range dependencies and improves prediction accuracy, achieving state-of-the-art results on hourly datasets.

Contribution

A novel two-stage framework explicitly models long-range structures and enhances forecast accuracy by integrating intermediate results with existing models.

Findings

Achieves state-of-the-art performance on M4 Hourly datasets.

Incorporating long-range structure improves forecast accuracy.

Combining neural networks with auto-regressive models captures linear and non-linear patterns.

Abstract

Seasonal time series Forecasting remains a challenging problem due to the long-term dependency from seasonality. In this paper, we propose a two-stage framework to forecast univariate seasonal time series. The first stage explicitly learns the long-range time series structure in a time window beyond the forecast horizon. By incorporating the learned long-range structure, the second stage can enhance the prediction accuracy in the forecast horizon. In both stages, we integrate the auto-regressive model with neural networks to capture both linear and non-linear characteristics in time series. Our framework achieves state-of-the-art performance on M4 Competition Hourly datasets. In particular, we show that incorporating the intermediate results generated in the first stage to existing forecast models can effectively enhance their prediction performance.