A Few-Shot LLM Framework for Extreme Day Classification in Electricity Markets
Saud Alghumayjan, Ming Yi, Bolun Xu
TL;DR
The paper tackles predicting next-day spikes in real-time electricity prices under limited labeled data. It introduces a few-shot LLM-based spike classifier that converts engineered daily system features into natural-language prompts and leverages embedding-based retrieval (FAISS) and Maximal Marginal Relevance to assemble informative few-shot exemplars, all without traditional model training. On ERCOT data, the LLM-based approach achieves performance competitive with SVM and XGBoost when ample historical data are available, and notably outperforms them in data-scarce scenarios, demonstrating data-efficient spike-day classification. This work suggests LLMs can support hedging and risk management in electricity markets by enabling robust extreme-event detection with limited labeled data.
Abstract
This paper proposes a few-shot classification framework based on Large Language Models (LLMs) to predict whether the next day will have spikes in real-time electricity prices. The approach aggregates system state information, including electricity demand, renewable generation, weather forecasts, and recent electricity prices, into a set of statistical features that are formatted as natural-language prompts and fed to an LLM along with general instructions. The model then determines the likelihood that the next day would be a spike day and reports a confidence score. Using historical data from the Texas electricity market, we demonstrate that this few-shot approach achieves performance comparable to supervised machine learning models, such as Support Vector Machines and XGBoost, and outperforms the latter two when limited historical data are available. These findings highlight the potential of LLMs as a data-efficient tool for classifying electricity price spikes in settings with scarce data.