SiamTST: A Novel Representation Learning Framework for Enhanced Multivariate Time Series Forecasting applied to Telco Networks

TL;DR

SiamTST is a new representation learning framework that combines Siamese networks, attention, and normalization to improve multivariate time series forecasting accuracy, especially in telecom networks.

Contribution

It introduces SiamTST, a novel framework integrating multiple techniques for better multivariate time series forecasting, with demonstrated superior performance on real-world data.

Findings

SiamTST outperforms existing methods in forecasting accuracy.

A simple linear network achieves competitive results, second only to SiamTST.

The framework is validated on real-world telecommunication data.

Abstract

We introduce SiamTST, a novel representation learning framework for multivariate time series. SiamTST integrates a Siamese network with attention, channel-independent patching, and normalization techniques to achieve superior performance. Evaluated on a real-world industrial telecommunication dataset, SiamTST demonstrates significant improvements in forecasting accuracy over existing methods. Notably, a simple linear network also shows competitive performance, achieving the second-best results, just behind SiamTST. The code is available at https://github.com/simenkristoff/SiamTST.