This study proposes a novel approach, intelligent text-analytic load forecasting (ITA-LF), to address the short-term load forecasting (STLF) problem by utilizing large language models (LLMs) and generative agents. This emphasizes the challenges faced by traditional forecasting methods in adapting to rapid changes and complex patterns in energy consumption, particularly during unexpected social events. It processes diverse unstructured data (e.g., historical loads, news, calendar dates, and weather), fine-tuning an LLM to enhance prediction accuracy and adaptability. An LLM-based agent with reasoning capabilities is introduced to select and understand relevant news, demonstrating the model’s ability to integrate diverse information for more precise forecasting. Our results surpass all baseline models in predictive accuracy, indicating that LLMs excel in managing the complexities of load forecasting patterns. This innovative approach not only improves forecasting accuracy but also indicates potential shifts in STLF paradigms by integrating unstructured data through advanced artificial intelligence (AI) techniques.
Wang et al. (2026) studied this question.