VGU RESEARCH REPOSITORY
Please use this identifier to cite or link to this item:
https://epub.vgu.edu.vn/handle/dlibvgu/1712| Title: | Domain adaptation for time series forecasting | Authors: | To Quang Huy | Keywords: | Time series forecasting;Domain adaptation | Issue Date: | 2023 | Abstract: | Time series forecasting is integral to various applications, from finance to healthcare, and the challenge often lies in the adaptability of prediction models across differing domains. This thesis delves into domain adaptation for time series forecasting, employing the innovative feature extractor, PatchTST. Four distinct methods are proposed and extensively analyzed: DANN (Domain-Adversarial Neural Network), DeepCORAL (Deep CORrelation ALignment), MMDA (Maximum Mean Discrepancy Adaptation), and RevIN (Reversible Instance Normalization). Each method offers different strategies to bridge the distribution gap between source and target domains. These experiments reveal that by leveraging the strength of PatchTST in feature extraction, combined with the domain adaptation capabilities of the proposed methods, substantial improvements in forecasting accuracy across diverse domains are achieved |
URI(1): | https://epub.vgu.edu.vn/handle/dlibvgu/1712 | Rights: | Attribution-NonCommercial 4.0 International |
| Appears in Collections: | Computer Science (CS) |
Files in This Item:
| File | Description | Size | Format | Existing users please Login |
|---|---|---|---|---|
| Domain adaptation for time series forecasting.pdf | 2.32 MB | Adobe PDF |
Page view(s)
168
checked on Nov 19, 2025
Download(s)
42
checked on Nov 19, 2025
Google ScholarTM
Check
This item is licensed under a Creative Commons License