GPT-2-Augmented Sequence Modeling for Short-Term Load Forecasting
ID：77
Submission ID：82 View Protection：ATTENDEE
Updated Time：2025-10-11 22:47:12 Hits：48
Poster Presentation
Start Time：2025-11-09 09:07 （Asia/Shanghai）
Duration：1min
Session：[P] Poster presentation » [P6] 6.AI-driven technology
No files
Abstract
Abstract -- Load forecasting serves as the foundation for power system operation and planning. Accurate load forecasting ensures the secure and reliable operation of power systems, reduces generation costs, and enhances economic efficiency. Recent studies demonstrate that large language models (LLMs) exhibit powerful capabilities in pattern recognition and reasoning for complex token sequences. The critical challenge lies in effectively aligning temporal patterns in time-series data with linguistic structures in natural language to leverage these capabilities. This paper proposes a large model-based time-series forecasting method for electrical load prediction. The approach leverages a pre-trained GPT-2 (Generative Pre-trained Transformer 2) model as its foundation while freezing parameters in its self-attention and feed-forward neural network layers. Fine-tuning is applied exclusively to the input embedding layer and output projection layer. Experimental results demonstrate that the proposed method achieves performance comparable to or superior against existing approaches across multiple electrical load forecasting tasks.
Keywords
Load forecasting, time-series data, LLM, GPT-2, Fine-tuning
Speaker
Comment submit