2025.6 – Present
GRAFT: Grid-Aware Load Forecasting with Multi-Source Textual Alignment and Fusion
Strict spatiotemporal alignment and cross-modal fusion between multi-source text and power-load signals.
A research framework for power-load forecasting that addresses the underuse of external information through strict spatiotemporal alignment, cross-modal fusion, and interpretability evaluation over multi-source text and load data. In February 2026, I submitted the first-author manuscript to Applied Energy, where it is currently under peer review.
01
Background
To address the insufficient use of external information in power-load forecasting, I built the GRAFT framework and carried out a systematic study on strict spatiotemporal alignment, cross-modal fusion, and interpretable evaluation over multi-source text and power-load data.
02
Core Work
- Built the dataset, designed the experiments, analysed the results, and wrote the manuscript
- Led the multi-source text processing pipeline and model training work
03
Methods
- Introduced text embeddings, cross-attention, and an external memory interface
- Built the study on a 2019–2021 multi-source text and load dataset covering five Australian states
- Established a three-dimensional evaluation framework over forecasting scale, region, and text source, with visualization analysis
04
Outcomes
- First-author manuscript: "GRAFT: Grid-Aware Load Forecasting with Multi-Source Textual Alignment and Fusion"
- Submitted to Applied Energy in February 2026
- Currently under peer review
05
Tech Stack
06
Figures
