Large Language Model Assisted Peer-to-Peer Energy Trading System

The peer-to-peer (P2P) energy trading market has been proposed as an effective mechanism to manage renewable energy sources. It enables users to transition from traditional consumers to prosumers and monetize their surplus energy to derive economic gains. Accurate day-ahead load forecasting for individual users is vital to improve the performance of P2P energy trading. However, the forecasting task remains challenging due to the high behavioral sensitivity of individual load profiles. In this article, a complete system is established, encompassing both the forecasting and the P2P trading model. The P2P market is coordinated by an energy trading coordinator based on the supply–demand ratio (SDR) pricing mechanism. A smoothing factor is introduced to address the strategy oscillations. To harness the contextual reasoning capabilities of large language models (LLMs), a two-stage prompt-assisted forecasting structure is developed. This structure enables the LLM to better understand the forecasting task by guiding it through detailed trend analysis to elucidate the rationale behind the load profiles. A multiround forecasting method with numerical postprocessing is designed, and the Gaussian kernel is used to reflect the density of data distribution. The case study indicates that for irregular load patterns of individual households, the proposed method achieves higher forecasting accuracy than the method using the long short-term memory, neural basis expansion analysis for time series forecasting, and Prophet models. Finally, a comprehensive analysis is conducted to explain how load forecasting errors affect load scheduling results under the SDR-based pricing mechanism in the P2P energy trading market.