Reproduce “DDT: A Dual-Masking Dual-Expert Transformer for Energy Time-Series Forecasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mingnan Zhu, Qixuan Zhang, Yixuan Cheng, Fangzhou Gu, Shiming Lin
A novel deep learning framework, DDT, is presented for energy time-series forecasting, integrating a dual-masking mechanism to balance causal consistency with adaptive feature selection and a dual-expert system for distinct temporal and cross-variable dependencies. The framework achieved superior performance across 8 diverse time-series datasets, consistently surpassing 10 state-of-the-art baselines.
Key insights
- The inherent conflict between maintaining causal consistency and enabling adaptive feature selection can be effectively resolved by synergistically combining static causal masks with data-driven dynamic masks.
- Decoupling and processing temporal dynamics and cross-variable correlations through a dual-expert system improves the model's ability to capture complex patterns in multi-variable time series.
- A comprehensive multi-level data processing pipeline, including noise detection, imputation, standardization, and manifold learning-based data augmentation, is crucial for robust performance on real-world energy time-series data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.